The multimedia communication needs high-performance, cost- effective communication techniques to transport data for the fast-growing multimedia traffic resulting from the recent deployment of World Wide Web (WWW), media-on-demand , and other multimedia applications. To transport a large volume, of multimedia data, high-performance servers are required to perform media processing and transfer. Typically, the high- performance multimedia server is a massively parallel processor with a high number of I/O ports, high storage capacity, fast signal processing, and excellent cost- performance. The parallel I/O ports of the server are connected to multiple clients through a network switch which uses parallel links in both switch-to-server and switch-to- client connections. In addition to media processing and storage, media communication is also a major function of the multimedia system. Without a high-performance communication network, a high-performance server can not deliver its full capacity of service to clients. Fortunately, there are many advanced communication technologies developed for networking, which can be adopted by the multimedia communication to economically deliver the full capacity of a high-performance multimedia service to clients. The VCSEL array technology has been developed for gigabit-rate parallel optical interconnections because of its high bandwidth, small-size, and easy-fabrication advantages. Several firms are developing multifiber, low-skew, low-cost ribbon cables to transfer signals form a VCSEL array. The OC12 SONET data-rate is widely used by high-performance multimedia communications for its high-data-rate and cost- effectiveness. Therefore, the OC12 VCSEL parallel optical interconnection is the ideal technology to meet the high- performance low-cost requirements for delivering affordable multimedia services to mass users. This paper describes a multimedia OC12 parallel optical interconnection using a VCSEL array transceiver, a multifiber ribbon cable, and MT connectors to achieve a high-performance, low-cost parallel link. A logical model of a multimedia server with parallel connections to an ATM switch, and to clients is presented. The design of the parallel optical link is analyzed. Furthermore, the link configured for testing, the test method, and test results are presented to confirm the analysis and to assure reliable link performance.

In this paper we continue our study of fast motion estimation techniques for H.263 video coding. We experiment with a hybrid method based on two techniques, the three-step search method, and out block-based gradient descent search method. Our method is a two phase procedure. For our first phase, borrowing from the three-step method, we do a search on a sparse grid. In the second phase, the block-based gradient descent search is applied to the candidate/s yielded from the first phase. The paper reports on the various parameters we have selected to test this new method, the timings for algorithms run using these parameters, and quality measurements for the resulting compressed video.

In digital libraries, documents are in digital forms and they are especially vulnerable from being copied. Existing copy detection methods exhaustively compare every single sentence of two documents and identify the degree of overlapping of the two documents. This approach is not scalable as the number of sentences for each document is often plentiful. In this paper, we propose a copy detection mechanism which could eliminate unnecessary comparisons. This is achieved by pre-parsing the documents to quantify their semantic meanings; comparisons between documents describing different topics could be eliminated s it will not serve any purpose to copy from a document describing an unrelated topic. This process is recursively applied to sections, subsections, subsubsections, etc. until we find two paragraphs which are highly related semantically. The paragraphs are then compared in detail, i.e., per-sentence basis, to determine if the paragraphs are overlapped in a substantive way. The parsing process is based on document retrieval techniques with some helpful heuristics that extract keywords from the documents to index the semantics for each document, section, subsection, and so forth. Weights based on relative occurrences of the keywords are assigned to individual keywords to form a keyword vector. The semantic relationships between different documents, sections, subsections, or paragraphs can be represented by the dot product of their corresponding keyword vectors as in document retrieval systems.

A new international standard for videoconferencing uses overlapped block motion compensation (OBMC) in the advanced prediction mode for improved motion compensation accuracy. Optimal motion estimates for OBMC may be computed by a computationally expensive iterative algorithm. A recent paper has suggested a computationally inexpensive two pass algorithm to compute motion estimates for OBMC. In the first pass, exhaustive search block matching is used to compute motion. This motion is then used for OBMC. Using a combination of the average OBMC error over and the variance of the error over a block and its neighbors, blocks are ordered. In the second pass, motion of the significant blocks form the ordering are optimized. In this paper, we investigate two fast search block motion estimation algorithms in the place of the exhaustive search algorithm. The algorithms are the three step search and gradient search block motion estimation algorithm. Simulations results with a head and shoulders video sequence reveals that the gradient search algorithm performs close to exhaustive search algorithm while the three step algorithm has lower performance.

Image warping concerns about transforming an image from one spatial coordinate to another. It is widely used for the vidual effect of deforming and morphing images in the film industry. A number of warping techniques have been introduced, which are mainly based on the corresponding pair mapping of feature points, feature vectors or feature patches (mostly triangular or quadrilateral). However, very often warping of an image object with an arbitrary shape is required. This requires a warping technique which is based on boundary contour instead of feature points or feature line-vectors. In addition, when feature point or feature vector based techniques are used, approximation of the object boundary by using point or vectors is required. In this case, the matching process of the corresponding pairs will be very time consuming if a fine approximation is required. In this paper, we propose a contour-based warping technique for warping image objects with arbitrary shapes. The novel idea of the new method is the introduction of mathematical morphology to allow a more flexible control of image warping. Two morphological operators are used as contour determinators. The erosion operator is used to warp image contents which are inside a user specified contour while the dilation operation is used to warp image contents which are outside of the contour. This new method is proposed to assist further development of a semi-automatic motion morphing system when accompanied with robust feature extractors such as deformable template or active contour model.

Computing histogram from images is an important step in generating feature vectors for content-based indexing of large image or video databases. In this paper, several methods for estimating histograms from transformed images are proposed. The results indicate that significant computation complexity reduction can be achieved while maintaining reasonable estimation accuracy.

Probabilistic relaxation is a powerful method for extracting features form images. Because the filtering process is basically independent of the relaxation process itself, probabilistic relaxation can be used to extract edges or ridges simply by choosing s suitable edge filter or line filter. Our recent work in hierarchical relaxation further improves the relaxation technique. Firstly, we use hierarchical constraints for the extraction of major features. Secondly, we partition the dictionary items according to the angle formed by the label sin each of the dictionary items to reduce the processing time for traversing the dictionary. The advantages of this hierarchical relaxation method are that it produces a more refined feature map and it improves the efficiency of the relaxation process by passing constraints from a low resolution relaxation process to a higher one. In this paper, we extend the idea of hierarchical relaxation to extracting 3D surfaces from volumetric data such as MRI data. Given a set of MRI images representing an object, we perform the relaxation process on each of the images to extract the contours of the object in the image. This relaxation process is constrained by the results of the same process applied to nearby images. A 3D geometric description of the object in the form of a polygon mesh can then be generated from the set of 2D contour curves. Results of the new method will also be demonstrated in this paper.

An optical WCHF-SDF (wavelet circular harmonic function synthetic discriminant function) filter is proposed. The proposed WCHF-SDF filter is synthesized by 2nd order harmonics of the four different wavelet transformed patterns by Haar wavelet function. This filter has full rotation and limited scale invariant properties. The scale invariant range can be increased by using more training patterns.

We are interested in the problem of classifying and segmenting bit streams with different source content and with different source coding in a communication channel. Although there are many researches on data segmentation, not much work is seen on this particular problem. Given zero and one observations of the bit stream, we first show that the windowed discrete FOurier transform enables us to distinguish fixed and variable length coded bit streams and in the case of fixed lengths coded bit stream, it can also determine the coding length of the bit stream. To further separate bit streams with variable length codes, we propose a classifier based on k-bit joint and conditional entropies. We present the joint and conditional entropy estimation schemes, and provide the upper bound for their performance. Then, we analyze the computational complexity of the entropy estimation. Finally experimental results are given to demonstrate the discriminant power of proposed entropy features.

This paper describes a new method of perceptual organization of thin networks using geometric properties. The key point of our approach is to consider perceptual organization as a problem of optimization: solutions to this problem are the best matchings between continuous curves and the low level primitives. First the quality of a grouping is defined with a class of functions related to the energy functions of active contours optimization. Such functions are computed recursively, and optimized from a local to a global level with an algorithm related to dynamic programming. This is followed by a selection procedure which rates and extracts principal groupings automatically and gives a new segmentation of image primitives, based on smoothed continuation. This segmentation is used to initialize a high level interpretation process involving projective reconstruction of 3D contours in sequences of images. The adaptability and robustness of this method have been tested on various situations, such as the extraction of ellipses from indoor scenes, roads from satellite pictures or blood vessels from medical images.

The comparison of video sequences at low bit-rates is object of an intense study. For high compression ratios, standard techniques produce noticeable artifacts in the reconstructed images. This article proposes a method for the post- processing of motion compensated hybrid coded video data. The technique uses a stochastic regularization approach which can be implemented with a simple and fast iterative algorithm. The approach has been applied to color video sequences produced by h.261 and h.263 standard compression methods. The post-processed sequences are aesthetically superior to the sequences obtained using the standard technique.

An active contour model has been widely applied to image segmentation and analysis. A contour model is able to be deformed to fit an expected object based on energy minimization. However, the conditions of initializing a contour model and defining internal energy parameters are important factors to the desired deformation process. We propose a new method to lessen these conditions so that the contour can be attracted correctly to the boundary of an object. With the dynamic programming approach, the contour model resolves energy ambiguity occurring in the image input. Then, the deformation process focusses on energy optimization that allows the model to maintain a desirable shape of an expected object. Fuzzy logic is applied to measure energy ambiguity and to form explicit shape knowledge that will guide the contour formation. The experimental results show that the contour model is able to make better decision in deforming itself in spite of an improper contour initialization and parameter definitions.

We propose in this paper a variable-coefficient fixed-length (VCFL) coding scheme for wavelet-based image transmission over noisy channels. When an image is transmitted through noisy channel with high throughput, both image compression and error-resistant coding scheme need to be considered. In this approach, an image is first decomposed into subbands by wavelet transform and quantized using an adaptive quantization scheme. The adaptive quantization is adaptive to both the frequency characteristics and the spatial constraints based on Gibbs random field. The traditional variable length entropy coding schemes, such as Huffman coding or arithmetic coding, and the fixed length coding such as LZW are usually very sensitive to channel noise for image transmission applications. Even with the insertion of synchronization symbols, they still cannot be directly employed without additional error correction/detection coding. To overcome the difficulty of image transmission over noisy channels, we propose to code the quantized subband coefficients with the VCFL scheme. This coding scheme attempts to keep the balance between redundancy removal, synchronization detection and error resilience. Part of the codebook is field based on the observation of the coefficient spatial distribution patterns in each subbands to alleviate the transmission of the codebook. The remaining code positions within the fixed length codebook can be utilized to combat channel errors by carefully arranging the code positions such that the codes with biggest transition cost will have the biggest Hamming distance. These positions can laos be filled with other frequently appeared coefficient composition sequences to achieve higher compression ratio. Experimental results of image transmission over noisy channels are reported to show the promising potential of the proposed coding scheme.

Malignant melanoma is the most common cancer in people less than 35 years of age and incident rates are increasing by approximately 5 percent per annum in many white populations, including British Columbia, Canada. In 1994, a clinical study has been established to digitize melanocytic lesions under a controlled environment. Lesions are digitized from patients who are referred to the Colored Pigment Lesion Clinic in the University of British Columbia. In this paper, we investigate how to use fractal dimensions (FDs) in measuring the irregularity of a skin lesion. In a previous project, we have experimented with 6 different methods to calculate fractal dimensions on a small number of images of skin lesions, and the simple box-counting method performed the best. However, the method did not exploit the intensity information of the images. With the new set of images which are digitized under the controlled environment, we utilize the differential box counting method to exploit such information. Four FD measures, including the direct FD, the horizontal and the vertical smoothing FDs, and the multi- fractal dimension of order two, are calculated based on the original color images. In addition, these 4 FD features are repeatedly calculate for the blue band of the images. This paper reports the different features through the calculations of the fractal dimensions and compares their differentiation power in the use of diagnosis of images of skin lesions.

Recently, there has been much research on video manipulation in the area of video-on-demand and video databases. Most of the recent work has so far focused on video classification, feature extraction, spatial reasoning and image retrieval; little work has been done on supporting adaptive video editing and production activities, nor has there been much work done on providing facilities for building a versatile video manipulation server. In this paper, we describe the development of an experimental video manipulation server called 'VIMS', which has been implemented at the Hong Kong University of Science and Technology. VIMS consists of two fundamental components: i) a video classification components (VCC) for the generation of effective indices necessary for structuring the video data, and ii) a conceptual clustering mechanism (CCM) having advanced object-oriented features and techniques. The former supports video structuring through camera break detection, shot classification using domain knowledge, as well as content-based retrieval through interactive learning, whereas the latter enables users to form, among other things, video programs form existing objects based on semantic features/index terms dynamically and adaptively. By tightly coupling CCM techniques together with VCC;s, the VIMS further allows the user to perform annotation-based and content-based retrieval in a well integrated and interleaved manner, which we regard as essential for a versatile video manipulation server. A prototype of VIMS embodying VCC and CCM has recently been constructed, running on the PC Pentium platform.

The application of multiresolution color quantization and indexing schemes to color-based image retrieval is investigated in this research. We first perform a thorough comparison of different quantization schemes in RGB, HSV, YUV and CIELUV color spaces. Then, a new feature based on the octree structure of color quanitization is proposed to achieve efficient multiresolution image indexing and retrieval. Extensive experiments are performed to illustrate the performance of the proposed multiresolution retrieval approach.

A high resolution computer-controlled CCD image capturing system is developed by using a 12 bits 1024 by 1024 pixels CCD camera and motorized RGB filters to grasp an image with color depth up to 36 bits. The filters distinguish the major components of color and collect them separately while the CCD camera maintains the spatial resolution and detector filling factor. The color separation can be done optically rather than electronically. The operation is simply by placing the capturing objects like color photos, slides and even x-ray transparencies under the camera system, the necessary parameters such as integration time, mixing level and light intensity are automatically adjusted by an on-line expert system. This greatly reduces the restrictions of the capturing species. This unique approach can save considerable time for adjusting the quality of image, give much more flexibility of manipulating captured object even if it is a 3D object with minimal setup fixers. In addition, cross sectional dimension of a 3D capturing object can be analyzed by adapting a fiber optic ring light source. It is particularly useful in non-contact metrology of a 3D structure. The digitized information can be stored in an easily transferable format. Users can also perform a special LUT mapping automatically or manually. Applications of the system include medical images archiving, printing quality control, 3D machine vision, and etc.

The error propagation caused by cell loss in MPEG video over ATM network may seriously deteriorate the video quality and reduce the error concealment effect. We present an efficient block interleaving and error concealment method for burst cell loss. At the transmitter, video information is interleaved and then packetized separately to reduce error damage. At the receiver, effective error concealment techniques are used for I, P, B frames respectively. Simulation results show satisfactory performance.

In multimedia applications, disk-arrays are often used to support he disk bandwidth requirements and data availability. Serious problems on available disk bandwidth and data availability could occur upon disk failure. In this paper, we discuss issues related to replicating data copies to provide high fault-tolerance required in a disk-array- based storage server. Replication and parity are two approaches to increase fault tolerance. By exploring theoretical aspects of these two concepts, we examine methods to build high availability disk arrays with multiple redundancy. The data allotment presented achieves the optimal fault-tolerance for goth triple and quadruple redundant disk arrays. Some theoretical properties of this problem are investigated.

A method to extract the information of a 3D object from the hologram is proposed with the application of the combination of computer and hologram. Simulated holograms of point object and 3D body-centered cubic lattice are made by overlapping the parallel reference coherent light. By means of FOurier optical analyze method, the objects were reconstructed successfully from the simulated hologram with object aberration less than 6.25 percent. FFT technology is used to quick the process.

CLLL TV technology is a new kind of night vision means. The vehicle and aircraft driving at night used in military and public security most suitably make full demonstration of its remarkable property. This thesis puts emphasis on studies on principle and technology of two-color false CLLL TV system. Relative technologies about realizing two-color false CLLL TV system are researched. After precisely theoretical calculation, a generalized 'white field balance' principle and its realizing technologies are raised. The part of experiment includes experiment design scheme, experiment results and analyses bout them. At last, several problems which need further studies to be solved are raised.

Multiple image registration plays a crucial role in multiple-sensor or multiple-spectral image fusion. We present an improved multi-resolutional SSDA that bases on wavelet decomposition. When computing the alignment parameters, we use a coarse-to-fine updating strategy and proper preprocessing stage. This approach yields sub-pixel accuracy to multi-spectral images with 2D geometric transformation of translations, rotations and scale changes. Experiment results show that our improved implement is robust and reliable for multispectral images.

The random flicker noise is the main character of low-light- level (LLL) image, so the noise suppressing is a key technique for LLL image processing. In this paper, the noise sources and its characteristics of LLL image are studied, and varied existing noise processing techniques are analyzed. On the basis of these, a novel filtering approach, space-time domain median filter, is presented. In this approach, the time correlation of LLL TV image is utilized and the time domain median filter is combined with the space domain median filter. Therefore the LLL image noise is reduced significantly, and the quality of LLL image is improved greatly.

Real-time low light level (LLL) image processing technology is the important developmental subject in the area of LLL night vision. But there is an essential distinction between the LLL TV image and ordinary TV image, so the conventional digital image processing technique aren't suitable for LLL image. In this paper, the noise theoretical model of LLL imaging system is described and the LLL image processing system is set up. With regard to the characteristics of LLL image and its noise, a novel noise suppression method, adaptive mode filter, is presented. The experimental results show that the adaptive mode filter can suppress the sharp noise of LLL image effectively, and as for the protection of the image edge, the property of adaptive mode filter is better that of median filter. Finally, the processing results and the conclusions are given.

A multispectral image fusion problem within the wavelet coefficient space has been studied in the present paper. In this algorithm, several kinds of area-based feature selection operators have been designed to select the image features in levels of the wavelet transform pyramids of source images. The problem of multiscale image enhancement and image fusion of different resolution has also been studied in the paper.

Recently the vector quantization (VQ) has received considerable interests as a powerful image data compression technique. However, studies of image coding with VQ have revealed that VQ for image compression suffers from edge degradation in the reproduced images. In this paper, we describe an adaptive learning method of the edge preserving VQ based on Kohonen's self-organizing feature map neural network. The learning procedure is performed by extracting the edge of the whole image, then adaptively adjusting the learning rate that are determined by the edge information of the image block. Compared with direct image VQ coding, the experiment results show the reproduced images quality are well improved, at the same compression ratio.

Fractal image compression allows fast decoding but suffers from long encoding time. By mapping similar regions within an image in the wavelet domain, we can use biorthogonal wavelet transform coefficients to find mappings which are identical to conventional blockwise mappings,thus the time consuming encoding process will improved by wavelet based hierarchical interpretation. We apply our methods to encode video sequence of digital image, in our scheme, the encoding process ID done for image pieces or motion vectors from the predecessor. The result shows a relatively low frame rate, since our target is transmitting video sequence in the level of requirements in ITU-T H.324 or the V.34 modem physical layer.

Using digital image processing based on computer aid processing technology, grey image can be pseudo-coloring coded and transformed to be a color image. Special data processing focus on some area of grey image such as grey photographic of x-ray can be made, it could greatly increase the detail resolution of a grey image and should be very useful in medicine and engineering. In this researching, we have paid more attention to small regions of image and similar joined areas and provided a useful technique as complementary coloring coding. Some new methods of signal picking up and contour recognition have been presented.

In this paper, on the basis of the analysis on the main noise sources of the Low Light Level (LLL) CCD TV system which our research laboratory developed, a new technique, interframe comparison denoise method, is put forward and taken to suppress the LLL image noise. The LLL image processing system and the LLL TV signal and noise test and analysis system are set up specially. Theory and experiment results show that interframe comparison denoise method is an effective LLL image processing technique. This time-domain image processing method is an improvement upon image frame- accumulation.

In this paper, a neural networks based approach for distortion invariant image recognition is presented. To reduce the dimension of the required networks, as well as to achieve invariancy, six distortion-invariant feature are extracted from each image and are used as inputs to the neural networks. These six features are derived from the modified geometrical moments of the image, which are calculated through a corrected discrete formula for computing moments more accurately. A multilayer perceptron network trained by the back-propagation algorithm can carry out the classification based on the above features. Experimental results on industrial tools and character recognition are to be given.

We have investigated an optical waveguide formed by aluminum nitride (AlN) thin film on sapphire. A good quality AlN thin film on sapphire substrate was prepared by metal organic chemical vapor deposition in this laboratory. A rutile prism coupler was employed to display the waveguide modes with the wavelength of 632.8, 532.1, 514.5 and 488.0nm. The refractive index and thickness of the waveguide material is obtained by prism-coupler measurement. The dispersion curve of AlN film is given and the dispersion equation is derived. The attenuation in waveguide is evaluated using scattering loss measurements by using a fiber probe in this experiment. The attenuation coefficient alpha is 1.5-2.1 cm^-1 various with different sample and the different modes of waveguide. The accuracy of the measurement is discussed.

In this paper, we present a fast algorithm that automatically detects human face regions in MPEG video sequences. The algorithm takes the DCT coefficients of MPEG video frames as the input, and output the positions of the bounding rectangles of the detected face regions. The algorithm is divided into three stages, where chrominance, shape, and DCT frequency information are used respectively. By detecting faces directly in the DCT domain, there is no need to carry out the inverse DCT transform, so that the algorithm runs in real time. In our experiments, the algorithm detected 92 percent of the faces in one hundred MPEG I-frames with few false alarms. The algorithm can be applied to JPEG still images or motion JPEG videos as well.

The purpose of this project is to develop a simplified DAVIC server on the Sun workstations under Unix and X window environment. DAVIC 1.0 is a comprehensive set of standards that define various types of end-to-end multi-media communication systems. More precisely, we implement only the high level server-client protocols and server service elements specified in DAVIC. This system can provide browsing service, download a file or a portion of it, and play back an MPEG sequence with VCR-like control. Limited by time, manpower and tools, not all the DAVIC specified elements are fully implemented. However, an implementation of a simple video server based on the DAVIC concept has been completed and demonstrated.

A developed binary image processing to detect the convolution interlaced minimum for recognizing and tracing the moving target is proposed in this paper. A diffraction device, which can divide an incidence light into 33 by 33 arrays evenly is designed and fabricated. A dual channel optical real-time processor with two 33 by 33 diffraction devices is built for parallel performing the recognition and tracing of the moving object. Some measures for compensating scale and rotation distortion for moving recognition are provided. The experimental results show that: a) The moving object can be recognized and traced parallel by one step of operation. b) The scale and rotation distortion of the moving target recognition is insensible. The change of rotation angle to be allowed in moving target recognition is from -25 degrees to +25 degrees.

This paper raises an algorithm of multisensor-multitarget data association based on the genetic algorithm for solving the problem of resolving and location to densely target in the system of multisensor data fusion. The experiment results indicate that this type of genetic algorithm has higher association success ratio, better adaptation and robustness to the change of type of sensors, resolving power and measurement error.

Correlation match algorithm is significant and distinctive, but it would be confronted with a critical problem in practical system. The problem is caused by sudden-variance of image. The purpose of this paper is to discuss the new methods to solve the problem through researching in the fields of different applications. One method is to rotate and reconstruct the new model map in laboratory situations. The other method is to utilize fixed features while image variance and symbolize them into the units of correlation match map.

Building a system for the users on the Internet to search interested resources or to provide guided browsing services has formed an important trend as a result of the rapid accumulation of information, the booming development of information system and the increasing requirement of resource sharing. In this paper, we present the Integrated resource Query and guidedBrowsing System (IQBS), which has been implemented at our laboratory, to achieve this goal. The scheme we proposed takes advantage of the strong power of the database system to structure and query data, and greatly promote the qua!ity and performance of searching resources in Internet. In addition, mechanisms are designed to guide the users on the Internet to conveniently browse resources with a reduced traffic cost. Keywords: Internet resource discovery, World wide web, Query processing, Rough set, Guided browsing, Supervised learning, Hypertext management.