PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.
This PDF file contains the front matter associated with SPIE
Proceedings Volume 7496, including the Title Page, Copyright
information, Table of Contents, and the Conference Committee listing.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Video networks is an emerging interdisciplinary field with significant and exciting scientific and technological
challenges. It has great promise in solving many real-world problems and enabling a broad range of applications,
including smart homes, video surveillance, environment and traffic monitoring, elderly care, intelligent
environments, and entertainment in public and private spaces. This paper provides an overview of the design of a
wireless video network as an experimental environment, camera selection, hand-off and control, anomaly detection.
It addresses challenging questions for individual identification using gait and face at a distance and present new
techniques and their comparison for robust identification.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
This paper presents a new approach for norm bounded continuous-time uncertain switch systems such as BTT vehicle.
Firstly, the flight trajectory has been divided into several intervals according to flight attitude. Meanwhile, the nominal
model could be obtained. Secondly, based on the linear matrix inequality (LMI), combining the guaranteed cost control
theory, the state feedback guaranteed cost control law is proposed. Then, to reduce the external disadvantage influence
on vehicle, the RBF-NN is used to compensate the intensive aerodynamic disturbance. Finally, taking BTT vehicle as the
research objective, the final results of the simulation are proved with effectiveness for the proposed design approach.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Selective visual attention can direct our gaze rapidly towards objects of interest in the view. Better coverage of target
region for attention can better serve for recognition. A novel method for evaluating how well the attended regions
contribute to the recognition of the target based on SIFT descriptors is proposed in this paper. The method is used to
evaluate the attended regions extracted by some visual attention mechanisms on real remote sensor images with different
geometric and photometric transformations and for different scene types. The evaluation method proposed in this paper
give an explicit, accurate and robust expression about the attended region in visual attention mechanisms and we believe
this method could be applicable in visual attention mechanisms in the future work.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
In a static monocular camera system, to gain a perfect 3D human body posture is a great challenge for Computer Vision
technology now. This paper presented human postures recognition from video sequences using the Quantum-Inspired
Immune Cloning Algorithm (QICA). The algorithm included three parts. Firstly, prior knowledge of human beings was
used, the key joint points of human could be detected automatically from the human contours and skeletons which could
be thinning from the contours; And due to the complexity of human movement, a forecasting mechanism of occlusion
joint points was addressed to get optimum 2D key joint points of human body; And then pose estimation recovered by
optimizing between the 2D projection of 3D human key joint points and 2D detection key joint points using QICA,
which recovered the movement of human body perfectly, because this algorithm could acquire not only the global
optimal solution, but the local optimal solution.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
We develop a novel approach for object detection and location task. This paper proposed a novel method to represent
local regions around keypoints, called lifetime. Lifetime of a keypoint is used to describe its stability. Together with
geometric relationships extractor, lifetime representations are embedded into a bag-of-features framework. The
framework has following properties. First, the keypoints are represented as the lifetime rather than vector-quantized.
Second, a simple and computationally efficient spatial pyramid structure is used to extract the geometric relationships
between the keypoints. We demonstrate the efficacy of the proposed approach on UIUC car dataset. The experimental
results showed that our approach has an excellent performance for object detection and localization.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
In this paper, a classification method of four moving objects including vehicle, human, motorcycle and bicycle in
surveillance video was presented by using machine learning idea. The method can be described as three steps: feature
selection, training of Support Vector Machine(SVM) classifier and performance evaluation. Firstly, a feature vector to
represent the discriminabilty of an object is described. From the profile of object, the ratio of width to height and
trisection ratio of width to height are firstly adopted as the distinct feature. Moreover, we use external rectangle to
approximate the object mask, which leads to a feature of rectangle degree standing for the ratio between the area of
object to the area of external rectangle. To cope with the invariance to scale, rotation and so on, Hu moment invariants,
Fourier descriptor and dispersedness were extracted as another kind of features. Secondly, a multi-class classifier were
designed based on two-class SVM. The idea behind the classifier structure is that the multi-class classification can be
converted to the combination of two-class classification. For our case, the final classification is the vote result of six twoclass
classifier. Thirdly, we determine the precise feature selection by experiments. According to the classification result,
we select different features for each two-class classifier. The true positive rate, false positive rate and discriminative
index are taken to evaluate the performance of the classifier. Experimental results show that the classifier achieves good
classification precision for the real and test data.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Due to vast growth of image databases, scene image classification methods have become increasingly important in
computer vision areas. We propose a new scene image classification framework based on combined feature and a latent
semantic model which is based on the Latent Dirichlet Allocation (LDA) in the statistical text literature. Here the model
is applied to visual words representation for images. We use Gibbs sampling for parameter estimation and use several
different numbers of topics at the same time to obtain the latent topic representation of images. We densely extract
multi-scale patches from images and get the combined feature on these patches. Our method is unsupervised. It can also
well represent semantic characteristic of images. We demonstrate the effectiveness of our approach by comparing it to
those used in previous work in this area. Experiments were conducted on three often used image databases, and our
method got better results than the others.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Currently, keyboards, mice, wands and joysticks are still the most popular interactive devices. While these devices
are mostly adequate, they are so unnatural that they are unable to give players the feeling of immersiveness.
Researchers have begun investigation into natural interfaces that are intuitively simple and unobtrusive to the
user. Recent advances in various signal-processing technologies, coupled with an explosion in the available
computing power, have given rise to a number of natural human computer interface (HCI) modalities: speech,
vision-based gesture recognition, etc. In this paper we propose a natural three dimensional (3D) game interface,
which uses the motion of the player fists in 3D space to achieve the control of sixd egree of freedom (DOFs). And
we also propose a real-time 3D fist tracking algorithm, which is based on stereo vision and Bayesian network.
Finally, a flying game is used to test our interface.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Ng-Jordan-Weiss (NJW) method is one of the most widely used spectral clustering algorithms. For a clustering problem
with K clusters, this method clusters data using the largest K eigenvectors of the normalized affinity matrix derived from
the data set. However, the top K eigenvectors are not always the most important eigenvectors for clustering. In this
paper, we propose an eigenvector selection method based on an ensemble of multiple eigenvector rankings (ESEER) for
spectral clustering. In ESEER method, first multiple rankings of eigenvectors are obtained by using the entropy metric,
which is used to measure the importance of each eigenvector, next the multiple eigenvector rankings are aggregated into
a single consensus one, then the first K eigenvectors in the consensus ranking list are adopted as the selected
eigenvectors. We have performed experiments on artificial data sets, standard data sets of UCI repository and
handwritten digits from MNIST database. The experimental results show that ESEER method is more effective than
NJW method in some cases.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Purpose: To generate a three-dimensional (3D) finite element (FE) model of human thorax which may provide the basis
of biomechanics simulation for the study of design effect and mechanism of safety belt when vehicle collision. Methods:
Using manually or semi-manually segmented method, the interested area can be segmented from the VCH (Visible
Chinese Human) dataset. The 3D surface model of thorax is visualized by using VTK (Visualization Toolkit) and further
translated into (Stereo Lithography) STL format, which approximates the geometry of solid model by representing the
boundaries with triangular facets. The data in STL format need to be normalized into NURBS surfaces and IGES format
using software such as Geomagic Studio to provide archetype for reverse engineering. The 3D FE model was established
using Ansys software. Results: The generated 3D FE model was an integrated thorax model which could reproduce
human's complicated structure morphology including clavicle, ribs, spine and sternum. It was consisted of 1 044 179
elements in total. Conclusions: Compared with the previous thorax model, this FE model enhanced the authenticity and
precision of results analysis obviously, which can provide a sound basis for analysis of human thorax biomechanical
research. Furthermore, using the method above, we can also establish 3D FE models of some other organizes and tissues
utilizing the VCH dataset.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
This paper presents a new affine registration approach for planar point pattern matching. A process of parameter space
clustering is implemented to confirm a one-to-one mapping between the maximal subsets of feature point sets in images.
For a best performance, a coarse parameter space and a fine parameter space are used for vectors comparison.
Experiments show that the method can produce positive results from a small number of feature points and intensive
noise.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
The development of digital aerial camera provides the possibility of acquiring highly overlapped aerial images with high
spatial resolution. In addition to its high spatial resolution that improves the capability of image interpretation, the highly
overlapped images provide favorable geometrical configuration with high redundancy. The high similarity of stereo
images is, thus, beneficial to the reliable image matching. Hence, the 3-D point clouds from the image matching have the
great potential in 3-D modeling. The topographic maps provide the distinct boundaries for building modeling. The
strategies of building reconstruction with existing topographic maps may improve the quality, cost, and efficiency for
building modeling. The objective of this investigation is to integrate highly overlapped aerial images and building
boundaries from topographic map. The proposed semi-automatic method includes 3-D lines extraction and polyhedral
model generation. In the beginning, the operator locates the initial lines by a graphical user interface. The initial lines are
refined by stereo images. The precise 3-D lines are processed into a 3-D modeling by an inference engine. The
experimental results indicate that the proposed method may reconstruct the 3-D building model effectively.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
This paper is a study, based on the limitation of human vision characteristic, of image recognition through the take
account of correction factor. Those aspects that have been explored focus on human eye modelings, including human
vision recognition characteristics and various mathematical modeling verify. By using Modulation Transfer Function
(MTF) curve evaluation recognition capability on the studied models, an optimum recognition model most compatible
to human eye physiology is summed up.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
The conventional display can show only one screen, but it is impossible to enlarge the size of a screen, for
example twice. Meanwhile the mirror supplies us with the same image but this mirror image is usually upside
down. Assume that the images on an original screen and a virtual screen in the mirror are completely different
and both images can be displayed independently. It would be possible to enlarge a screen area twice. This
extension method enables the observers to show the virtual image plane and to enlarge a screen area twice.
Although the displaying region is doubled, this virtual display could not produce 3D images. In this paper, we
present an extension method using a unidirectional diffusing image screen and an improvement for displaying a
3D image using orthogonal polarized image projection.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
In going from two-class to multi-class classification, most boosting algorithms have been restricted to reducing multiclass
problem to multiple two-class problems. In the paper, a direct multi-class AdaBoost algorithm is adopted to face
recognition. Then the weighted classification trees are extended from stumps as weak learners to fulfill the multi-class
learning. The multi-class boosting algorithm has the following features: A K-class classification problem is treated
simultaneously without reducing it to multiple binary classification problems; only one lost function per iteration is fitted;
the algorithmic structure is compact and easy to implement. The experimental results both on UCI dataset and YaleA
face dataset show the meanings of the proposed algorithm.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
In spite of years of research there is tremendous scope for improvement in face recognition systems. There is continuing
requirement for better face recognition systems with lesser computational complexity and higher accuracy. In this paper
a new algorithm is proposed which addresses these requirements by integrating the computationally efficient technique
of Fisher faces with Super-Resolution. The algorithm helps in identifying a person from sub-pixel shifted low resolution
facial images of those people whose high resolution images are present in our library. The focus of this paper is only on
face recognition and not on face detection and extraction. Extensive testing of the proposed algorithm was performed on
images from the Essex university Faces94 database. In our test many parameters such as the number of low resolution
input images and registration error etc. were varied and their effect on the accuracy (i.e. percentage of correct results)
and throughput of the face recognition system was studied. It is shown that this method has high accuracy and low
computational complexity and that it is robust to Gaussian blur, and salt and pepper noises due to camera and errors in
image registration through experimental results performed on Faces94 database.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
A new hybrid registration approach that combines hierarchically structured quadrilateral displacement estimating and
local optical flow is proposed in this paper, for the purpose of super resolution (SR) reconstruction for visual surveillance
application. The proposed registration approach for complicated motions circumstance consists of two steps: a
hierarchical quadrilateral displacement estimation algorithm is designed to get coarse motion estimation as initial
prediction; then a local optical flow method is employed to obtain more accurate motion estimation for each pixel within
every matched quadrilateral. A ROI-based SR construction scheme using the proposed registration approach is presented
for iterative reconstruction of region of interest in the scene. Experimental results show that significant improvements are
achieved by applying our method than using previous methods, which suggests the effectiveness of the proposed method.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Support vector machines (SVMs) have become useful and universal learning machines. SVMs construct a decision
function by support vectors (SVs) and their corresponding weights. The training phase of SVMs definitely uses all
training samples, which leads to a large computational complexity for a large scale sample set. Moreover support vectors
could not be found until a quadratic programming (QP) problem is solved. Actually we know only SVs play a role in the
decision function. Hence, pseudo density estimation (PDE) is presented to extract a set of boundary vectors (BVs) which
may contain SVs. The PDE method is a variant of Parzen window method. Hyperspheres are considered as the window
functions. In our method, for each sample we construct a hypersphere with an unfixed radius. The ratio of the number of
samples contained in the hypersphere of a sample to the total training samples can be taken as the pseudo density of the
corresponding sample. The set of BVs is taken as the training input to SVMs. In doing so, it speeds the training
procedure of SVMs. It is convenient for PDE to determine its parameter. The experiments show that SVMs using PDE
have the similar generalization performance to SVMs.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Research on the generation of natural phenomena has many applications in special effects of movie, battlefield
simulation and virtual reality, etc. Based on video synthesis technique, a new approach is proposed for the synthesis of
natural phenomena, including flowing water and fire flame. From the fire and flow video, the seamless video of arbitrary
length is generated. Then, the interaction between wind and fire flame is achieved through the skeleton of flame. Later,
the flow is also synthesized by extending the video textures using an edge resample method. Finally, we can integrate the
synthesized natural phenomena into a virtual scene.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Extraction of high-resolution face image is crucial to detect suspect from low-resolution surveillance videos. Though
previously published super-resolution image reconstruction techniques could produce a qualified high-resolution image
from a set of simulated low-resolution images, but the reconstructed image from real low-resolution videos is always
blurring. Two main reasons contribute for this: the process of image registration is ill-posed in nature and the sub-pixel
information provided by the real video sequences is far less sufficient. In this paper, a joint image registration and face
pattern-based high-resolution image reconstruction algorithm was proposed to tackle these two problems. Experimental
results are also provided to demonstrate the effectiveness of the proposed algorithm.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
To overcome the difficulty of reconstruction for small object, the paper addresses a series of improved methods. The
reconstruction mainly includes: orientation, measurement and modeling. The multi-view matching based on refining TIN
can combine the process of automatic measuring and modeling. Based on coarse-to-fine tragedy, it can increase
effectively precision and efficiency. The experiment results demonstrate that these approaches are effective and
applicable to reconstruction of small archeology with sufficient texture features.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
The relevance vector machine is sparse model in the Bayesian framework, its mathematics model doesn't have
regularization coefficient and its kernel functions don't need to satisfy Mercer's condition. RVM present the good
generalization performance, and its predictions are probabilistic. In this paper, a hyperspectral imagery classification
method based on the relevance machine is brought forward. We introduce the sparse Bayesian classification model,
regard the RVM learning as the maximization of marginal likelihood, and select the fast sequential sparse Bayesian
learning algorithm. Through the experiment of PHI imagery classification, the advantages of the relevance machine used
in hyperspectral imagery classification are given out.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Facial expressions play important role in human communication. The understanding of facial expression is a basic
requirement in the development of next generation human computer interaction systems. Researches show that the
intrinsic facial features always hide in low dimensional facial subspaces. This paper presents facial parts based facial
expression recognition system with sparse representation classifier. Sparse representation classifier exploits sparse
representation to select face features and classify facial expressions. The sparse solution is obtained by solving l1 -norm
minimization problem with constraint of linear combination equation. Experimental results show that sparse
representation is efficient for facial expression recognition and sparse representation classifier obtain much higher
recognition accuracies than other compared methods.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
This paper presents a new approach of Remotely Sensed data classification based on Variable Precision Rough
set(VPRS) and BP neural network, compared to traditional rough sets, VPRS is more robust to noise and can generate
more concise and representative classification rules of the remote sensing image. After the rules are deduced, they are
fed to the BP neural network, which results in short training time and a high classification accuracy.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
This paper describes how to construct a hyper-graph model from a large corpus of multi-view images using local
invariant features. We commence by representing each image with a graph, which is constructed from a group of
selected SIFT features. We then propose a new pairwise clustering method based on a graph matching similarity
measure. The positive example graphs of a specific class accompanied with a set of negative example graphs are
clustered into one or more clusters, which minimize an entropy function with a restriction defined on the
F-measure( 2/(1recall+1/ precision) ). Each cluster is implified into a tree structure composed of a series of irreducible graphs,
and for each of which a node co-occurrence probability matrix is obtained. Finally, a recognition oriented class specific
hyper-graph(CSHG) is automatically generated from the given graph set. Experiments are performed on over 50K
training images spanning ~500 objects and over 20K test images of 68 objects. This demonstrates the scalability and
recognition performance of our model.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
It is necessary to reconstruct a large-scale landing-site mapping by recovering and registering the local scenes into a
uniform annular scene for planetary exploration missions. This paper proposed a global relax iterative optimization
method to registering the local scenes into a uniform annular scene. For this scheme, the transform matrix between any
two adjacent 3D local scenes is fitted based on Carley transform. Subsequently, these local 3D scenes are registered into
a uniform coordinate system using relax iterative optimization method. This optimization method has been tested on the
image sequence of outdoor scenes. Experimental results show that the global registration means error decreases
significantly from 1.33 meters to 0.002 meters in 47 images.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Independent component analysis (ICA) provides an efficient approach to characterizing higher-order statistical
relationships in texture images. For the classification of textures based on ICA, a fundamental problem lies on the
selection of ICA features which are desired to maximize the separability between classes. In this paper, the efficiency of
various ICA features for texture classification is investigated, which involves ICA coefficients and their various statistics
with a focus on the higher-order statistics to take into account the non-Gaussian property of ICA coefficients. By
evaluating the ICA features on the classification of twenty-five classes of Brodatz texture images, it has been shown that
the higher-order statistics of ICA coefficients offer efficient discrimination of textures and the combination of variance,
skewness and kurtosis is a better alternative to the previously reported ICA features. By comparing the performance of
ICA features to their principal component analysis (PCA) counterparts, it is further revealed that the advantage of ICA
for texture classification can be obtained by using the higher-order statistics of ICA coefficients.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
A novel object tracking algorithm for FLIR imagery based on mean shift using multiple features is proposed to improve
the tracking performance. First, the appearance model of infrared object is represented in the combination of gray space,
LBP texture space, and orientation space with different feature weight. And then, the mean shift algorithm is employed to
find the object location. An on-line feature weight update mechanism is developed based on Fisher criteria, which measure
the discrimination of object and background effectively. Experiment results demonstrate the effectiveness and robustness
of the proposed method for object tracking in FLIR imagery.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
This paper focus on TV news programs and design a content-based news video browsing and retrieval system, NVRS,
which is convenient for users to fast browsing and retrieving news video by different categories such as political,
finance, amusement, etc. Combining audiovisual features and caption text information, the system automatically
segments a complete news program into separate news stories. NVRS supports keyword-based news story retrieval,
category-based news story browsing and generates key-frame-based video abstract for each story. Experiments show
that the method of story segmentation is effective and the retrieval is also efficient.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
We present a novel online face recognition approach for video stream in this paper. Our method includes two stages:
pre-training and online training. In the pre-training phase, our method observes interactions, collects batches of input
data, and attempts to estimate their distributions (Box-Cox transformation is adopted here to normalize rough estimates).
In the online training phase, our method incrementally improves classifiers' knowledge of the face space and updates it
continuously with incremental eigenspace analysis. The performance achieved by our method shows its great potential in
video stream processing.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Inspired by the idiotypic network theory, a new artificial immune network classifier for SAR image is proposed in this
paper. In the proposed algorithm, only one B-cell instead of many B-cells is used to denote a class so as to reduce the
scale of network as well as avoid the suppression operation between B-cells; moreover, a new affinity function based on
the correct rate is proposed to realize antigen priority based the evaluation strategy. The proposed algorithm has been
extensively compared with Fuzzy C-means (FCM), Multiple-Valued Immune Network algorithm (MVIN), and Clonal
Selection Algorithm for classifier (CSA) over two SAR images. The result of experiment indicates the superiority of the
algorithm over FCM, MVIN and CSA on classification accuracy and robustness.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
We propose a new distance estimation technique by boosting and apply it to enhance the effectiveness of classifier when
the training set is insufficient. The proposed method is called Boosted Distance based on local and global dissimilarity
representation (BDLGDR). It is a modified method of Boosted Distance. Rather than simply differentiating the feature
vectors, we calculate a new dissimilarity representation of each couple of feature vectors. This new dissimilarity
representation contains two parts: local dissimilarity representation part and global dissimilarity representation part. The
proposed method does not only achieve high classification accuracy when the training set is insufficient but when the
number of training set is sufficient it also can achieve as high accuracy as AdaBoost. The method has been thoroughly
tested on several databases of high-resolution (1.25m) Terra-SAR images. In the first experiment, we decreased the
number of the training sample per class from 10 to 1. The result showed that the proposed method outperformed both
Boosted Distance and AdaBoost. In the second experiment, we used sufficient training samples. The experimental result
illuminated that the proposed method performed at least as well as AdaBoost and needed fewer iteration rounds to
converge than Boosted Distance.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
This paper proposes an ant colony optimization (ACO) based approach for point pattern matching (PPM) under affine
transformation. In the paper, the point sets matching problem is formulated as a mixed variable (binary and continuous)
optimization problem. The ACO is used to search for the optimal transformation parameters. There are two contributions
made in this paper. Firstly, we manage to modify the original ACO method by combining it with the leastsquares
method. Thus, it can handle with the continuous spatial mapping parameters searching. Secondly, we introduce a
threshold to correspondence finding, which rejects outliers and enhances veracity while using "Nearest Neighbors
Search". Experiments demonstrate the validity and robustness of the algorithm.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
A novel fuzzy approach for the detection of traffic signs in natural environments is presented. More than 3000 road
images were collected under different weather conditions by a digital camera, and used for testing this approach. Every
RGB image was converted into HSV colour space, and segmented by the hue and saturation thresholds. A symmetrical
detector was used to extract the local features of the regions of interest (ROI), and the shape of ROI was determined by a
fuzzy shape recognizer which invoked a set of fuzzy rules. The experimental results show that the proposed algorithm is
translation, rotation and scaling invariant, and gives reliable shape recognition in complex traffic scenes where clustering
and partial occlusion normally occur.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Autonomous service on orbit is a new developing trend of space service, in order to realize on orbit servicing, the
problem of autonomous relative navigation needs to be solved. On the foundation of several typical schemes used by
American, Russian as well as ESA etc., the main conception of our autonomous video navigation system is determined.
The beacon system is composed of five beacon lamps, and has four invariant features relative to transformations of
rotation, translation and scale. Two sets of double camera systems constituted of three fixed lens cameras with short
focus are applied to the longer distance and the shorter distance respectively. Three kinds of measures are put forward to
suppress the interference of miscellaneous lights. Then, the study and simulation of beacon recognition as well as the
determination of relative positions and attitudes are expounded in this paper, the algorithm flow chart and corresponding
simulation results are given. A kind of spiral capturing algorithm is used to increase the capturing efficiency, and the
beacon recognition algorithm is designed according to the invariant features of beacon system to increase the recognition
capability. In the determination of relative positions and attitudes, monocular algorithm and binocular algorithm are
combined to ensure the reliability. Simulation results have verified the feasibility of the design of autonomous video
navigation sensor and the autonomous video navigation techniques.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Celestial spectrum recognition is an indispensable part of any workable automated data processing system of
celestial objects. Many methods have been proposed for spectra recognition, in which most of them concerned about
feature extraction. In this paper, we present a Bayesian classifier based on Kernel Density Estimation (KDE) which
is composed of the following two steps: In the first step, linear Principle Component Analysis (PCA) is used to
extract features to decrease computational complexity and make the distribution of spectral data more compact and
useful for classification. In the second step, namely classification step, KDE and Expectation Maximum (EM)
algorithm are used to estimate class conditional density and the bandwidth of kernel function respectively. The
experimental results show that the proposed method can achieve satisfactory performance over the real observational
data of Sloan Digital Sky Survey (SDSS).
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
This paper developed two learning procedure, respectively, based on the orthogonal least squares (OLS) method and
the "Innovation-Contribution" criterion (ICc) proposed newly. The orthogonal use of the stepwise-regression algorithm
of the ICc mages the model structure independent of the selected term sequence and reduces the cluster region further as
compared with orthogonal least squares (OLS). as the Bayesian information criteria (BIC) method is incorporate into the
clustering process of the ICc, except for the widths of Gaussian functions, it has no other parameter that need tuning ,but
the user is required to specify the tolerance ρ, which is relevant to noises and will be difficult to implement in the real
system, for the OLS algorithm. The two algorithms are employed to the Radial Basis Function Neural Networks
(RBFNN) to compare its performance for different noise nonlinear dynamic systems. Experimental results show that
they provide an efficient approximation to the required results for fitting models, but the clustering procedures of the ICc
is substantially better solutions than does the OLS.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
A fast and flexible technique of 3D modeling of building based on image sequence has been proposed in this paper. It
firstly describes the importance of the study in this area, then gives a detailed analysis of each step of the whole
reconstruction process, including homonymy points selection, determination of key points' relationship, the solution
method of splicing among points appeared on different stereo images, texture mapping and so on. Finally, real data has
been used to validate the proposed technique, using VC++6.0 and OpenGL to realize the visualization of the buildings
artificial interactively, and satisfied results have been obtained demonstrating the effectiveness and flexibility of the
approach.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
In this paper, a center matching scheme is proposed for constructing a consensus function in the k-means cluster
ensemble learning. Each k-means clusterer outputs a sequence with k cluster centers. We randomly select a cluster center
sequence as a reference one, and then we rearrange the other cluster center sequences according to the reference
sequence. Then we label the data using these matched cluster center sequences. Hence we get multiple partitions or
clusterings. Finally, multiple clusterings are combined to the best labeling by using combination rules, such as the
majority voting rule, the weighted voting rule and the selective weighted voting rule. Experimental results on 7 UCI data
sets show that our ensemble methods could improve the clustering results effectively.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
In this paper, we first analyzed the possible change of support vector set after new samples are added, then presented a
new support vector machine incremental learning algorithm. This algorithm reconstructed SVM classifier through the
selection of training samples in incremental learning based on change regularity of support vectors after new samples are
added. Finally, the algorithm has a higher classification accuracy than traditional SVM incremental algorithms through
experimental verification.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
The human vision system has visual functions for viewing 3D images with a correct depth. These functions are
called accommodation, vergence and binocular stereopsis. Most 3D display system utilizes binocular stereopsis.
The authors have developed a monocular 3D vision system with accommodation mechanism, which is useful
function for perceiving depth.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Since SVM is very sensitive to outliers and noises in the training set and the fuzzy feature exists in remote sensing
images, we hereby studied fuzzy support vector machine based on the affinity among samples. The fuzzy membership is
defined by not only the relation between a sample and its cluster center, but also the affinity among samples. A method
defining the affinity among samples is proposed using a sphere with minimum volume while containing maximum of the
samples. Then, the fuzzy membership is defined according to the position of samples in sphere space, which
distinguished between the valid samples and the outliers or noises. The experiment results show, it discriminates support
vectors with noise or outliers much better. Experimental results show that our method performs better than SVM in
classification of the images in Wuhan and with less influnence by the noise interference.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
We present a new approach to face detection with skin color mixture models and asymmetric AdaBoost. First, non-skin
color pixels of the input image are rapidly removed based on skin color mixture models in RGB and YCbCr
chrominance spaces, from which we extract candidate face regions. Then, face detection with fast asymmetric AdaBoost
is carried out in candidate face regions where ratios of pixels of skin color to non-skin color are beyond certain
thresholds. To further reduce the computational cost, the integral image technique is employed to calculate ratios of
pixels of skin color to non-skin color in candidate face regions. Finally, false alarms are gradually merged and removed
by relative geometric relation and the rate of skin color pixels on the intersection line of candidate face regions.
Experimental results show that our proposed method reduces significantly false alarms and the processing time while
achieves detection rates of more than 99%.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Eye corner detection is important for eye extraction, face normalization, other facial landmark extraction and so on. We
present a feature-based method for eye corner detection from static images in this paper. This method is capable of
locating eye corners automatically. The process of eye corner detection is divided into two stages: classifier training and
classifier application. For training, two classifiers trained by AdaBoost with Haar-like features, are skillfully designed to
detect inner eye corners and outer eye corners. Then, two classifiers are applied to input images to search targets. Eye
corners are finally located according to two eye models from targets. Experimental results tested on BioID face database
and our own database demonstrate that our method obtains a high accuracy under clutter conditions.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
For the special shape of the tunnel and the limitation to the image sequences capturing inside the tunnel, conventional
stereo algorithms are difficult to apply for 3D reconstruction of the tunnel. This paper presents an automatic
reconstruction method for 3D tunnel reconstruction from the monocular image. The proposed method assumes one 3D
wireframe model of the tunnel can be easily obtained by a few known data. Any lines on this model can be expressed
with Plucker matrix, by using detected lines on the image, the line projection matrix can be obtained. Then the whole 3D
wireframe model can be mapped onto the image, so that a texture can be extracted for the 3D wireframe model, then one
textured 3D tunnel model is obtained. The experimental results show that our method can be easily and effective
performed in the practice.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
In this paper, we present a new member of the biometrics family, i.e. nose pores, based on DLPP. Little work has been
done on nose pores as a biometric identifier. In this work, we made use of a database of nose pore images obtained over
a long period to examine the performance of nose pores as a biometric identifier. First, the midpoint and midline were
located and taken as reference for the ROI segmentation after nose image was segmented. Second, nose pore feature was
filtered by LOG filters. Third, the extracted pore was projected to low dimensional space by DLPP. Finally, the feature
in low dimension was classified by Euclidean distance. This research showed that the nose pore is a promising candidate
for biometric identification and deserves further research. The experimental results based on the unique nose pores
database demonstrated that nose pores can give a 91.91% correct recognition rate for biometric identification, which
showed this biometric identifier's feasibility and effectiveness. Compared with result without using DLPP, the feature
extraction by DLPP was more precise.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
The recent increase of space threats yields the idea of using the existing star trackers to perform surveillance of space
objects from space. In the missions, due to the observer attitude dynamics, smearing affects the observed stars on the
image in space surveillance. Besides, the reflecting flying space objects or debris as spurious stars affects the attitude
determination. These are devastating for most star identification algorithms in star trackers. To resolve the problems, this
paper defines a star pattern, called Flower code, which is composed of angular distances and circular angles as the
characteristics of the pivot star. The angular distances are used for initial lookup table match. Moreover, the circular
angles are used for the cyclic dynamic match between the sensor pattern and the pattern candidates from the initial
match. The focus of the results is the evaluation of the influence of the reflecting flying spacecraft or debris as spurious
stars and the attitude dynamics of the observer spacecraft, on the performance of the algorithms. A number of
experiments are carried out on simulated images. The results demonstrated that the proposed method is efficient and
robust.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
A new nonlinear control strategy incorporated the dynamic inversion method with wavelet neural networks is presented
for the nonlinear coupling system of Bank-to-Turn(BTT) missile in reentry phase. The basic control law is designed by
using the dynamic inversion feedback linearization method, and the online learning wavelet neural network is used to
compensate the inversion error due to aerodynamic parameter errors, modeling imprecise and external disturbance in
view of the time-frequency localization properties of wavelet transform. Weights adjusting laws are derived according to
Lyapunov stability theory, which can guarantee the boundedness of all signals in the whole system. Furthermore, robust
stability of the closed-loop system under this tracking law is proved. Finally, the six degree-of-freedom(6DOF)
simulation results have shown that the attitude angles can track the anticipant command precisely under the
circumstances of existing external disturbance and in the presence of parameter uncertainty. It means that the
dependence on model by dynamic inversion method is reduced and the robustness of control system is enhanced by
using wavelet neural network(WNN) to reconstruct inversion error on-line.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
An airborne vehicle must avoid obstacles like towers, fences, tree branches, mountains and building across the flight path.
So the ability to detect and locate obstacles using on-board sensors is an essential step in the autonomous navigation of
aircraft low-altitude flight. In this paper, a novel passive range method using conditional random field (CRF) is presented
to map the 3D scene in front of a moving aircraft with image sequences obtained from a forward-looking imaging sensor.
Finally, An dynamic graph cuts method was presented for the CRF model to recursively update thedepth map.
Experimental data demonstrates the effectiveness of our approach.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
MLESAC is one of the most widely used robust estimators in the field of computer vision. A shortcoming of this method
is its low efficiency. An enhancement of MLESAC, the locally optimized MLESAC (LO-MLESAC) is proposed.
LO-MLESAC adopts the same sample strategy and likelihood theory as the previous approach and an additional
generalized model optimization step is applied to the models with the best quality. Results are given for several image
sequences. It is demonstrated that this method gives results superior to original MLESAC.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
This paper studies statistics based scene change detection in the video streaming scenario, and three
scene feature metrics including histogram distance, chi-square distance, and Bhattacharyya distance have been
investigated. With the unique characteristics of triangular inequality and non-singularity, Bhattacharyya distance
has been proposed as a viable scene change metrics. It outperforms much better than the other two in that it
calculates and maximizes the feature vector distance between multi-modal clusters in a hyper-sphere space. The
experiments are conducted and the precision recall statistics are compared, and the results support our analysis.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
With the advent of information age, especially with the rapid development of network, "information explosion"
problem has emerged. How to improve the classifier's training precision steadily with accumulation of the samples is the
original idea of the incremental learning. Support Vector Machine (SVM) has been successfully applied in many pattern
recognition fields. While its complex computation is the bottle-neck to deal with large-scale data. It's important to do
researches on the SVM's incremental learning. This article proposes a SVM's incremental learning algorithm based on
the filtering fixed partition of the data set. This article firstly presents "Two-class problem"s algorithm and then
generalizes it to the "Multiclass problem" algorithm by the One-vs-One method. The experimental results on three types
of data sets' classification show that the proposed incremental learning technique can greatly improve the efficiency of
SVM learning. SVM Incremental learning can not only ensure the correct identification rate but also speedup the training
process.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
The traditional classification method based on the spectral information in pixel level faces the problem of spectrum
confusion. Pixels on multi-temporal image show dependencies in both the spatial and temporal domains besides spectral
information. When spectral information has limited discriminative power, spatial-temporal dependencies can help to
remove the spectral confusion. Two ETM+ images in different seasons after processing are used and supervised
classification algorithm-the maximum likelihood classification (MLC) is used to initialize the algorithm proposed in this
article. Then a Markov Random Fields (MRF) model is used to model the spatial-temporal contextual prior probabilities
of images. Lastly the likelihood estimates of spectral observation from MLC and conditional spatial-temporal priors from
MRF are integrated into posterior estimates by Bayes rule, the optimal classification was achieved when the
classification corresponds to maximum a posteriori (MAP). The results show that MRF is an efficient probabilistic model
for analysis of spatial and temporal contextual information. A spatial-temporal classification algorithm that explicitly
integrates spectral, spatial and temporal information in multi-temporal images can achieve significant improvements over
non-contextual classification. Some errors have been avoided because of the integration of space and time information.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Since manual surface reconstruction is very costly and time consuming, the development of automatic algorithm is of
great importance. In this paper a fully automated technique based on hierarchical structure analysis of the building to
extract urban building models from LIDAR data is presented. In the processing of reconstruction, the existing automatic
algorithm can solve some simple building reconstructions, such as flat roof, gabled roof. As to complex buildings, many
researchers use external information or manual interaction for help because of the complexity of the reconstruction and
the uncertainty of the building models especially in urban areas. The contour has the characteristics of closed loop, not
intersect and deterministic topological relationship, which can be used to extract building ROI (region of interesting). A
contours tree is constructed, the topological relationships between the different contours which extracted by TIN from
the LIDAR data are established, then the relationships among each hierarchical model can be determined by the analysis
of the topological relationship among contour clusters and a component tree corresponding to the building can be
constructed by tracing the contours tree. The accurate edges of hierarchical model can be gained by the "polarized
cornerity index"-based polygonal approximation of the contour. Especially, a 3D model recognition based on 2D shape
recognition is employed. According to the characteristics of the contours, the category of the primitive parts can be
classified. We assemble the hierarchical models by using the topological relationships among layers, then, the complete
model of the building can be obtained. Experimental results show that the proposed algorithm is suitable for
automatically producing building models including most complex buildings from LIDAR data in urban areas.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Digital human modeling and skin deformation is the most challenging and widely used computer animation technology.
In all existing methods the skinning algorithm has many advantages such as fast implement speed and simple operating
process, but it also has the flaws such as local 'collapse' and 'drape' phenomena. An improved skinning method is
presented that can effectively reduce traditional flaws of this method. After the key terminologies are introduced the
improvement for skinning deformation is illustrated in detail. The main measures include that (1) adding extra joints on
JSL to minimize the distance between adjacent joints and thus joint's importance attribute is introduced, (2) using joint
cluster to replace single joint, and (3) creating the corresponding relationship between skin and JSLs based on flexible
model and multi-joints-binding method (MJBM), that is to say binding one skin vertex to several joints using distance
criterion and weight coefficient is the function of distances between the skin vertex to its related joints. All these
improvements can make the skin deformation more smoothing. In order to verify the validity of the method we
implement the algorithm above proposed using Visual C++ 6.0 and OpenGL language, and the digital hand finger
bending and leg kicking deformation are achieved. The experimental results show that the skin deformation is very
natural and meets the habit of human self.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
In this paper, we consider the combined problem of distinguishing classes from the background and from each other, and
propose an improved framework based on the previous state-of-the-art approaches. In the process of building ECOC
(Error Correcting Output Coding) matrix (also called as sharing matrix), we adopt an encoding rule of one-versus-all,
and maximize Hamming distance in categories as far as possible through heuristic search in sharing-code maps (i.e.,
layer joint boosting). Then the final classifier is responsible for detection, and ECOC matrix for recognition. In order to
make full use of the output of the final classifier and its corresponding ECOC matrix, the following measures are worth
considering: Firstly, a logistic function of the output mentioned above is used for a posterior probability of each
codeword. Therefore the identified class label is the one corresponding to the codeword of Maximum a posteriori
(MAP). Secondly, a similarity measurement utilizing the confusion matrix is advanced to focus on the similarities
between classes. Thirdly, for the purpose of adaptive adjustment in Hamming distance, we change the subsequent search
coding method according to the confusion matrix until the training errors are convergent. The experimental results
illustrate the effectiveness of the proposed approach.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Diverse modeling frameworks have been utilized with the ultimate goal of translating brain cortical signals into
prediction of visible behavior. The inputs to these models are usually multidimensional neural recordings collected from
relevant regions of a monkey's brain while the outputs are the associated behavior which is typically the 2-D or 3-D
hand position of a primate. Here our task is to set up a proper model in order to figure out the move trajectories by input
the neural signals which are simultaneously collected in the experiment. In this paper, we propose to use Echo State
Networks (ESN) to map the neural firing activities into hand positions. ESN is a newly developed recurrent neural
network(RNN) model. Besides its dynamic property and short term memory just as other recurrent neural networks have,
it has a special echo state property which endows it with the ability to model nonlinear dynamic systems powerfully.
What distinguished it from transitional recurrent neural networks most significantly is its special learning method. In this
paper we train this net with a refined version of its typical training method and get a better model.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
A method combined Kernel Principal Component Analysis (KPCA) with BP neural network is proposed for
multispectral remote sensing image classification in this paper. Firstly, the KPCA transformation including Gaussian
KPCA and polynomial KPCA is carried out to get the former three uncorrelated bands containing most information of
the TM images with seven bands. Secondly, BP neural network classification is executed using the three bands data after
KPCA transformation. For testifying, both the classical PCA and the KPCA are applied to the multispectral Landsat TM
data for feature extraction. The results demonstrate that the method proposed in this paper can improve the classification
accuracy compared with that of principal component analysis (PCA) and BP neural network.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Using the kernel trick idea and the kernels as features idea, we can construct two kinds of nonlinear feature spaces,
where linear feature extraction algorithms can be employed to extract nonlinear features. Thus, we have two approaches
to transform an existing linear feature extraction algorithm into its nonlinear counterpart. It has been proved that they are
equivalent up to different scalings on each feature by rigorous theoretical analysis. In this paper, we perform experiments
on several benchmark datasets and give a comparative study of the two kernel ideas applied to certain feature extraction
algorithms such as linear discriminant analysis and principal component analysis. These results provide a better
understanding of the kernel method.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
The goal for vehicle based multi-sensors integrated system is extracting spatial data and attribute data from mass data of
different sources. The main sensors used in vehicle based multi-sensors integrated system for surveying and mapping are
multi-pair of stereo cameras. There are two types of calibrations for stereo cameras in multi-sensors integrated system.
One is relative calibration, and the other is absolute calibration. In this paper, a full mathematical model for stereo
cameras on multi-sensors integrated system has been described and the mathematical models for translate coordinates
from stereo camera sub-system to geographical coordinate system are also given out. And integrated calibration of stereo
cameras has been realized successfully. With the calibration result, the total precision of the system used in this paper is
is suitable for medium or small scale surveying and mapping in urban area.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Stereo correspondence methods rely on matching costs for computing the similarity of image locations. In this paper, a
stereo matching cost based on orthogonal Gaussian-Hermite moments (OGHMs) space is proposed and evaluated
together with different matching costs with respect to radiometric variations of the input images. Their performance is
measured in the presence of global intensity changes, local intensity changes, and image noise. Using existing stereo
datasets with ground-truth disparities taken under controlled changes of exposure and lighting, the experiment shows the
novel matching cost method guarantee a more insensitive result against different kinds of radiometric differences.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
There are a large number of fuzzy concepts and fuzzy phenomena in traditional Chinese medicine, which have led to
great difficulties for study of traditional Chinese medicine. In this paper, the mathematical methods are used to quantify
fuzzy concepts of drugs and prescription. We put forward the process of innovation formulations and selection method in
Chinese medicine based on the Possibility Construction Space Theory (PCST) and fuzzy pattern recognition.
Experimental results show that the method of selecting medicines from a number of characteristics of traditional Chinese
medicine is consistent with the basic theory of traditional Chinese medicine. The results also reflect the integrated effects
of the innovation compound. Through the use of the innovation formulations system, we expect to provide software tools
for developing new traditional Chinese medicine and to inspire traditional Chinese medicine researchers to develop
novel drugs.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Traditional camera lens calibrations need a 3D physical model with many control points on it, the coordination of these
points are measured high precision. These points should be evenly distributed in space for calibration precision, and it is
difficult to make the model especially big model. In this paper, a calibration plate was made and the plate was moved in
its normal direction, parallel, no rotation, it is implemented to provide large area 3D control points with variable Z values,
the images of the plate were made in different places. the moving plate is a 3D model, moreover, it could increase
control points significant, the plate is much easier to make than traditional 3D physical model, the plate could be big one,
the calibration parameters of camera lens were calculated by different control points group, get the mean value of these
parameters. This method is proved high precision, stabilization and robustness. Experiments show such an approach is
effective for reconstructing 3D objects in computer vision system.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
In this paper, a novel method for infrared face recognition based on blood perfusion is proposed in this paper. Firstly,
thermal images are converted into blood perfusion domain to enlarge between-class distance and lessen within-class
distance, which makes full use of the biological feature of the human face. Based on the ratio of between-class distance
to within-class distance (Ratio of Distance (RD)) in sub-blocks, block-PCA is utilized to get the local discrimination
information, which can solve the small sample size problem (the null space problem). Finally, The FLD is applied to the
holistic features combined by the extracted coefficients from the information of all sub-blocks. The experiments illustrate
that the block-PCA+FLD doesn't discard the useful discriminant information in the holistic characters and the method
proposed in this paper has better performance compared with traditional methods.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Incremental learning is an efficient scheme for reducing computational complexity of batch learning. Label information
in each update is helpful to update discriminative model in incremental learning. However, the procedure of labeling
samples is always a time-consuming and tedious task. In this paper, we propose two labeling algorithms for unknown
samples, one is discriminative Transductive Confidence Machine for K-Nearest Neighbor (TCM-KNN), the other is its
improved algorithm for choosing good quality discriminative samples and enhancing the performance of the procedure
of labeling samples; and then these methods is applied in the incremental learning[2] before updating model. Experiment
on PIE database has been carried out for comparing their recognition rate and complexity. Extensive experimental results
show that the proposed method for incremental learning is more robust and effective than batch learning.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
In this paper, a general and efficient facial feature extraction approach, global search linear discriminant analysis
(GSLDA), is presented. It is designed to solve the puzzle of standard linear discriminant analysis (LDA) for small
sample size problems (SSSP). Compared with PCA-LDA, in GSLDA, raw data dimension can be greatly decreased
without discarding important discriminant information. In this process, all basis vectors of the non-null eigen-space of
the scatter matrix is worked out, and then the well-known global search strategy, genetic algorithm, is enrolled to select
basis vectors to construct a new feature space which has optimal discriminant ability. In contrast with PCA, this
approach reserves more information for recognition. Therefore, this process enhances the performance of LDA for SSSP,
and eventually the recognition performance. This strategy has been tested on the ORL and Yale face database.
Experiment results show that this approach works much better than classical facial feature extraction methods.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Many algorithms have been proposed in literature for feature selection; unfortunately, none of them ensures a perfect
result. Here we propose an adaptive sequential floating forward feature selection algorithm which achieves accuracy
results higher than that of already existing algorithms and naturally adaptive for implementation into the number of best
feature subset to be selected. The basic idea of the proposed algorithm is to adopt two relatively well-settled algorithms
for the problem at hand and combine mutual information and Cross-Validation through suitable fusion techniques, with
the aim of taking advantage of the adopted algorithms' capabilities, at the same time, limiting their deficiencies. This
method adaptively obtains the number of features to be selected according to dimensions of original feature set, and
Dempster-Shafer Evidential Theory is used to fuse Max-Relevance, Min-Redundancy and CVFS. Extensive experiments
show that the higher accuracy of classification and the less redundancy of features could be achieved.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
In this paper, we are concentrating on the passive geometric camera calibration, which calculates the
geometric parameters of all cameras in a camera system based on certain predefined object. The to-becalculated
parameters include intrinsic parameters (e.g. focal length and aspect ratio) and extrinsic
parameters (camera 3D position and orientation). Geometric camera calibration is the first and fundamental
step in e.g. 3D measurement and 3D reconstruction systems. Its accuracy always determines the quality of
those systems. Based on deliberately built calibration object, passive camera calibration can generate highly
accurate parameter approximation of the camera system. However, past research mainly focused on either
single or stereo camera systems, few have studied calibration of multi-camera and special camera systems
(e.g. mirror based cameras). In this paper, we will introduce a coherence systematic calibration system that
is fully automatic, workable under difficult situations and applicable for single, stereo, and multiple normal
or special camera systems.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
In this paper, a novel local manifold spectral clustering with fuzzy c-means (FCM) data condensation is presented.
Firstly, a multilayer FCM data condensation method is performed on the original data to contain a condensation subset.
Secondly, the spectral clustering algorithm based on the local manifold distance measure is used to realize the
classification of the condensation subset. Finally, the nearest neighbor method is adopted to obtain the clustering result
of the original data. Compared with the standard spectral clustering algorithm, the novel method is more robust and has
the advantages of effectively dealing with the large scale data. In our experiments, we first analyze the performances of
multilayer FCM data condensation and local manifold distance measure, then apply our method to solve image
segmentation and the large Brodatz texture images classification. The experimental results show that the method is
effective and extensible, and especially the runtime of this method is acceptable.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Infrared image offers the main advantage over visible image of being invariant to illumination changes for face
recognition. In this paper, based on the introduction of main methods of linear subspace analysis, such as Principal
Component Analysis (PCA) , Linear Discriminant Analysis(LDA) and Fast Independent Component Analysis
(FastICA),the application of these methods to the recognition of infrared face images offered by OTCBVS workshop are
investigated, and the advantages and disadvantages are compared. Experimental results show that the combination
approach of PCA and LDA leads to better classification performance than single PCA approach or LDA approach, while
the FastICA approach leads to the best classification performance with the improvement of nearly 5% compared with the
combination approach.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
This article proposes a novel method for object tracking using active basis model. Active basis model has the
strong competitive ability of object representation, especially the ability of modeling the sketch variables of the
image. And each basis element of active basis model has a certain freedom because it can shift its location or
orientation or both within a little range, then such tuning of basis elements can adjust to the variation of the
object of interest from frame to frame. We have done some experiments on the video sets which just have only
one or two moving objects in each frame for avoiding any other interference, in evidence, the tracker based on
active basis model can perfectly capture the object of interest accurately from frame to frame, despite of part
occlusion. The disadvantage of out method is that the computational complexity of Gabor wavelet is high, so
this method cannot implement the real-time tracking currently.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Super-resolution reconstruction technology is to explore new information between the under-sampling image series
obtained from the same scene and to achieve the high-resolution picture through image fusion in sub-pixel level. The
traditional super-resolution fusion methods for sub-sampling images need motion estimation and motion interpolation
and construct multi-resolution pyramid to obtain high-resolution, yet the function of the human beings' visual features
are ignored. In this paper, a novel resolution reconstruction for under-sampling images of static scene based on the
human vision model is considered by introducing PCNN (Pulse Coupled Neural Network) model, which simplifies and
improves the input model, internal behavior and control parameters selection. The proposed super-resolution image
fusion algorithm based on PCNN-wavelet is aimed at the down-sampling image series in a static scene. And on the basis
of keeping the original features, we introduce Relief Filter(RF) to the control and judge segment to overcome the effect
of random factors(such as noise, etc) effectively to achieve the aim that highlighting interested object though the fusion.
Numerical simulations show that the new algorithm has the better performance in retaining more details and keeping
high resolution.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
This paper studies the optical diffraction, defocus, and sampling effects on the high speed imaging
system. Through detailed analysis, we formulate the point spread function (PSF) and optical transfer function (OTF)
which characterize the optical system in the spatial and frequency domain respectively. With the approximate priori
knowledge of their PSF, we propose two integer focus operators and analyze their OTFs in association with those
of Laplacian and the optimal focus operator. The experimental results and comparison validate our theoretical
investigation.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
This paper presents an integrated bottom-up and top-down computing process for parsing cars. By parsing, it means
detecting all instances in an input image, and aligning their constituent parts, if appeared. The output of parsing is to
construct configurations of car instances. In real scenarios such as in street scenes, cars often appear with different
degree of occlusions, which bring two problems in car parsing: (1) Occlusions often fail those holistic methods, so we
use a deformable part-based model. In terms of generative models, this paper proposed a star-like pictorial structural
model based on the active basis model. The presented model is hierarchical and deformable. (2) In turn, part-based
models entail integrated bottom-up and top-down computing processes. Bottom-up processes generated hypotheses from
input images for each node in the deformable model. Top-down processes are followed to verify those bottom-up
hypotheses in terms of their configurations. In order to evaluate the proposed method, we build up a dataset in which
different kinds of occlusions are randomly added to cars. Experiment results show that the integrated bottom-up and topdown
process improves the performance greatly.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
The relevance vector machine (RVM) is a sparse regression kernel model. It not only generates a much sparser model
but provides better generalization performance than the standard support vector machine (SVM). Relevance vectors and
support vectors are both selected from the input vector set. This may limit model flexibility. Recently, we propose
Relevance Units Machine (RUM). RUM treats relevance units (RUs) as part of the parameters of the model. However,
the number of RUs must be selected before using RUM. In this paper, we use Akaike's Information Criterion (AIC) to
select the number of the RUs. The experiment results show that based on AIC RUM maintains all the advantages of
RVM and offers superior sparsity.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
SOPC (System on Programmable Chip) is an on-chip programmable system based on large scale Field Programmable
Arrays (FPGAs). This paper presented an implementation of an SOPC system with a custom hardware neural network
using Altera FPGA chip-EP2C35F672C. The embedded Nios processor was used as the test bench. The test result
showed that the SOPC Platform with hardware neural network is faster than the software implementation respectively
and the accuracy of the design meets the requirement of system. The verified SOPC system can closely model real-world
system, which will have wide applications in different areas such as pattern recognition, data mining and signal
processing.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Image matching is a fundamental aspect of many problems in computer vision. We describe a novel wide baseline
matching method based on scale invariant feature descriptor. First, corners in image pairs are detected based on an
improved Curvature Scale-Space (CSS) technique. These corners are relatively invariant to affine transformations, and
are represented by using Scale Invariant Feature Transform (SIFT) descriptor to provide robust matching. The nearest
neighbor distance is then applied to remove mismatched corners. Finally, the robust estimation algorithm, RANSAC, is
adopt to estimate the fundamental matrix from the correspondence, and at the same time identify inlying matches.
Experiments demonstrate the feasibility of this method.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Combining multiple clusterers is emerged as a powerful method for improving both the robustness and the stability of
unsupervised classification solutions. In this paper, k-means selective cluster ensembles based on multiple feature subsets
are proposed. In the ensemble, a random subset of features is used to train an individual k-means clusterer. In the final
step, the selective weighted voting scheme is used for finding the best partition. The consensus function is constructed by
relabeling all partitions of clusterers and finding the best partition. Experimental results on 4 UCI data sets show that our
ensemble method can improve the clustering performance.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
The Scale Invariant Feature Transform (SIFT) proposed by David G. Lowe has been used in face recognition and proved
to perform well. Recently, a new detector and descriptor, named Speed-Up Robust Features (SURF) suggested by
Herbert Bay, attracts people's attentions. SURF is a scale and in-plane rotation invariant detector and descriptor with
comparable or even better performance with SIFT. Because each of SURF feature has only 64 dimensions in general and
an indexing scheme is built by using the sign of the Laplacian, SURF is much faster than the 128-dimensional SIFT at
the matching step. Thus based on the above advantages of SURF, we propose to exploit SURF features in face
recognition in this paper.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Firstly, the concepts of discernibility degree and relative discernibility degree are presented based on general binary
relations. Then the properties of these concepts are analyzed. Furthermore, an efficient attribute reduction algorithm is
designed based on the relative discernibility degree. Especially, the attribute reduction algorithm is able to deal with
various kinds of extended models of classical rough set theory, such as the tolerance relation-based rough set model,
non-symmetric similarity relation-based rough set model. Finally, the theoretical analysis is backed up with numerical
examples to prove that the proposed reduction method is an effective technique to select useful features and eliminate
redundant and irrelevant information.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
A new ortho updating method by means of the aerial triangulation which Ground Control Points (GCPs) are auto
matched from the existing ortho photos and DEM has been introduced in the paper. In this new method, a two steps automatching
arithmetic have been used: the one is used in auto matching between the adjacent aerial photos in order to auto
transferring tie points of the block; the other is used to matching the tie points with the existing ortho photos by the aid of
DEM, if the matching succeed, the ground coordinates of the tie points could be obtained and then it could be used as a
GCP in the aerial triangulation. Therefore, the ortho updating method presented in the paper could not only extract tie
points, but also could extract enormous GCPs to calculate the Exterior Orientation parameters (EOP) of the aerial images
with high accuracy and high reliability.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
In this paper, an object extraction algorithm from complex scenes is presented. Firstly, Textons are modeled by the joint
distribution of filter responses. This distribution is represented by Texton (cluster centre) frequencies. Secondly,
classification of a novel image proceeds by mapping the image to a Texton distribution and comparing this distribution
to the learnt models. So the detection of possible object regions is performed. During the verification stage, the
knowledge about the scene and the geometry of the objects is represented by means of t graph, and especially, the
knowledge about the surrounding of the object is used in order to support the detection of individual objects. Finally,
Bayes nets are selected to verify those possible objects as a useful tool. The test on the dataset in building scenes shows
that the proposed algorithm has a better performance, compared with the similar methods.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Point inclusion testing is to test the relationship of a fixed point and a polygon, an algorithm for this problem is presented.
A ray was drawn from the fixed point, the ray may have crossing point with the edge of the polygon, two vectors were
given from the fixed point to the endpoints of the edge which has crossing point with the ray, then calculate their vector
product. Two functions were defined according to the vector product, the decision for point inclusion testing was based
on this two functions. calculate the vector product for each edge that has crossing point with the ray from the fixed point,
If the amount of positive vector product and negative vector product are the same, the point is outside the polygon;
otherwise, the point is inside the polygon. This method can decrease computing time and can avoid some mistakes of
other algorithm, and the main part of algorithm only needs 5n times subtraction, 2n times multiplication, n times
*judgement, its algorithm complexity is 0(n). this algorithm is suitable for some other cases including self-intersection
polygon and polygon with hole.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
This paper presents one from-Coarse-to-Fine approach to geo-register photogrammetry image sequence to un-textured
LIDAR point clouds. First, Photogrammetric Automatic Aerial Triangulation (AAT) Technique is altered to coarsely
geo-register image sequence to LIDAR points; After that, LIDAR-Image-Mutual line feature extraction and
correspondence strategy is proposed to generate 3D space line from LIDAR points, extract image lines and establish
correspondence effectively; Next, "Line Photogrammetric" technique is utilized to map generated conjunctive line
features and to refine previous geo-register result under Least Squares adjustment; Finally, geo-registration experiment
with developed approach is conducted and contributions are concluded as well. The experimental result shows that
proposed approach is robust and effective.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
The goal of single-frame Super-Resolution is to improve the spatial resolution of a given low-resolution image. However,
it is ill-posed. Regularization which can be interpreted as the way of finding the prior distribution of images plays a
crucial role in solving this problem. Example-based approach is one of the well-established regularization techniques for
image process based on the prior information stored in the database, which is also used for image Super-Resolution
reconstruction. This paper previews the Exampled-based Super-Resolution approach which is based on Freeman's work.
We show how the example images to be used to generate training set, describing the Super-Resolution synthesis
processing based on the training set, with the plausible experiment results on the single-frame image scale-up. Finally,
the related problems and future challenges in this field are also mentioned.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
A new SAR image classification method is proposed based on the remarkable correlation of feature channels, which are
obtained by dividing the Fourier plane into sectors according to the frequency and direction. Compared with other
feature correlation based methods in wavelet and Contourlet domain, our method operates directly in the Fourier plane,
which is more robust with regard to images size and more flexible in the capture of frequency and directional
information. Moreover, the use of FFT transform is computationally more attractive. Experimental results on the Brodatz
textures and SAR images demonstrate the effectiveness and efficiency of the proposed method.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
We present a vision-based landing algorithm for an autonomous helicopter under complex environment (there are several
suspected targets). The algorithm is integrated with algorithms for visual acquisition, recognition of the target and
computing the navigation information. In our algorithm, we use international standard landing mark as our landing target.
The experiment results demonstrate that our algorithm has the feature of robustness, accuracy and real time. It can meet
the actual flight requirements well: the average processing time of a 640×480 image is less than 40ms; the position error
is below 5cm in each axis of translation; the angle error is below 3.5°. Based on the algorithm, we win the champion of
the aerial robot competition in the 2008 China robot competition and the RoboCup China open.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
In this paper, the artificial neural network (ANN) algorithm for solving numerical differentiation problem is
considered and some numerical tests are given to demonstrate the effectiveness of our results.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Graph cut criterion has been proven to be robust and applicable in clustering problems. In this paper the graph cut
criterion is applied to construct a supervised dimensionality reduction. A new graph cut, scaling cut, is proposed based on
the classical normalized cut. Scaling cut depicts the relationship between samples, which makes it can handle the
heteroscedastic and multimodel data in which LDA fails. Meanwhile, the solution to scaling cut is global optimal for it is
a generalized eigenvalue problem. To obtain a more reasonable projection matrix and reduce the computational
complexity as well, the localized k-nearest neighbor graph is introduced in, which leads to equivalent or better results
compared with scaling cut.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
A buried 3-D optical waveguide have been fabricated by Tl+-Na+ ion-exchange technology on BK7 glass substrates, the
refractive-index profile of the waveguides in cross-sectional were determined with the interference method technique;
The refractive-index distribution function n(x,y) for optical waveguides in cross-sectional was improved, and the
refractive index profile was reconstructed by MATLAB, and the shapes of the index profile in the cross-sectional are in
good agreement with experimental results; The shapes of the refractive-index distribution of 3-D optical waveguide in
cross-sectional can be reconstructed by the improved function n(x,y) easily.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
The image information change law with the change of scale parameter is first studied in terms of regarding ramp edge
and step edge as measurement of image information. Then the paper proposes a kind of adaptive recursive algorithm of
scale parameter with the module of visual characters. The information of the image is uniformly distributed among each
layer in the algorithm. The method can avoid the problem of complicated computation or over- distortion because of
losing too much key information. The experimental results show that the uniformly distributed information is more
reasonable for human to perceive an image which will be useful for higher-level image processing technologies such as
object recognition.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Thanks to the development of network, video devices are used in various aspects of our lives such as video conference,
video call, etc. It is, however, very common that people tend to look at the computer screen instead of the video capture
device, which can affect the communication. We present a novel approach to solve this problem. Through estimating the
divert head angle and eyeball location recognition, we can estimate the visual angle. By moving the eyeball to adjust
visual line, it looks like that we're looking at each other. Our work is on the basis of face alignment and there is a
geometric 3D model and novel R -α, K -β relationship analysis methods adopted in this paper.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
A new 3D automatic reconstruction method of micro solid of revolution is presented in this paper. In the implementation
procedure of this method, image sequence of the solid of revolution of 360° is obtained, which rotation speed is
controlled by motor precisely, in the rotate photographic mode of back light. Firstly, we need calibrate the height of
turntable, the size of pixel and rotation axis of turntable. Then according to the calibration result of rotation axis, the
height of turntable, rotation angle and the pixel size, the contour points of each image can be transformed into 3D points
in the reference coordinate system to generate the point cloud model. Finally, the surface geometrical model of solid of
revolution is obtained by using the relationship of two adjacent contours. Experimental results on real images are
presented, which demonstrate the effectiveness of the Approach.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
The method of extracting the trunk by Gray-Level Co-occurrence matrix method has been proposed in this paper,
meanwhile, it contrasted with the effects of region growing algorithm based on seed point which was selected from the
stem base; then the information of diameter at breast height, diameter of main branch and branch height was figured out
according to photogrammetry study . Experimental results show that the methods can effectively segment the trunk and
the background, furthermore, it's more convenient to obtain the standing tree features automatically and accurately.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
We proposed a novel method to extract and reduce Category Specific SIFT Descriptor (CSSD). Our approach is based
on two facts. One is that in many images there are always more than two different objects and this brings on ambiguity
of the categories which they should belong to. The other is that the number of SIFT features for one image often varies
from tens to thousands, matching these SIFT features of two arbitrary images brings high computational costs. As for the
first fact, those category specific SIFT features hide among the sum of SIFT information, we aim to filter out the
contributive SFIT information to category recognition by clustering all. With respect to the second fact, the SIFT clusters
can instruct to reduce each image SIFT features by keeping high occurrence frequency ones. So the more precious and
smaller SIFT features depending on its category specific features can be obtained. Another main highlight of our
approach is the sensible use of affinity propagation to address the definition of clustering category K more objectively.
Extensive experiments shows that the RCSSD (Reduced CSSD) obtained by affinity propagation clustering outperforms
the original SIFT descriptor and RCSSD by using K-means approach.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Ear recognition is a kind of the novel representative subjects in the field of non-disturbance biometrics authentication
and is becoming received wide attention in academic research. In this paper, the ear recognition problem based on
texture analysis is discussed. A novel local wavelet binary pattern descriptor combining local binary pattern descriptor
with wavelet transform filter is presented. And an ear recognition approach based on local wavelet binary pattern
descriptor and support vector machines classification is proposed, which is tested on USTB ear image set. The
experiment results show that the ear recognition scheme using local feature descriptor and transform filter is effective
and promising. The performance of support vector machines classifier is better than that of K Nearest Neighbor classifier.
The best combination occurs under the Chi square distance and 'reverse biorthogonal 3.1' wavelet, and the 96.86%
cross- validation recognition rate is obtained.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Automatic feature matching, especially region matching, has made great progress in recent years, and a great deal of
descriptor-based methods have been proposed. However, when constructing these descriptors for irregular regions, an
extra step of fitting the irregular regions into fixed shape must be implemented in advance. The fitting step can cause
great errors, and thus may result in poor matching. The main purpose of this paper is developing a strategy of
constructing descriptors for irregular regions without any extra fitting steps. In this paper, two groups of descriptors are
developed: one is the gradient-based descriptors and the other is the Harris-based descriptors. Experiments show that
descriptors proposed in this paper can perform great and robust for irregular region matching on real images.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
A method for real-time fast detection and symptom analysis of cracks in highway asphalt pavement is proposed. At the
first step, the fissure characteristic of the acquired and de-noised pavement images is analyzed and then extracted by the
method based on gray value comparison between the current pixel and its neighboring pixels, the false cracks are deleted
by using the computed measurement of crack features, thus the true fissures are detected. The most important step is the
symptom analysis of the cracks in the pavement image, all the data could be analyzed and be the basis for the agencies to
remedy and manage the pavement. Quantities of images are processed and the results show that the proposed method can
detect the pavement distress information actually, and has robustness and availability.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
In view of the crucial deficiency of the traditional diagnosis approaches for photoelectric tracking devices and the output
of more sufficient diagnosis information, in this paper, an virtual fault diagnosis system based on open graphic
library(OpenGL) is proposed. Firstly, some interrelated key principles and technology of virtual reality, visualization and
intelligent fault diagnosis technology are put forward. Then, the demand analysis and architecture of the system are
elaborated. Next, details of interrelated essential implementation issues are also discussed, including the the 3D modeling
of the related diagnosis equipments, key development process and design via OpenGL. Practical applications and
experiments illuminate that the proposed approach is feasible and effective.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
In this paper, we study an integrated framework to
generate expressive sketch animation from real video with
user interaction. It consists of two mainly steps: (i) image
sketch computation by a learning-based edge detector; (ii)
temporal sketch propagation by a robust stochastic
matching algorithm. In the first step, given a video clip,
the edge probability map on each frame is first computed
by a discriminative model that is trained with a collection
of various features. A template sketch is flexibly extracted
from the beginning frame by threshold tuning, where user
intervention is allowed to perfect the sketch template.
Then this template is matched and localized to the
following image sketches over frames by the graph-based
matching algorithm. User interaction is allowed to
sequentially correct the matching results. A number of
sketch animations from real videos are presented to verify
this framework in the experiments.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.