This PDF file contains the editorial “Guest Editorial: Performance Evaluation of Signal and Image Processing Systems” for OE Vol. 30 Issue 02

A historical perspective on the evolution of performance evaluation technology in the field of automatic target recognition (ATR) systems is presented. ATR systems have been under development for the last 30 years, and each generation of ATRs and their predecessors have been accompanied by their corresponding stages in the evolution of performance evaluation technology. This perspective will shed light on this field and will answer the questions where are we now and where are we heading?

Automatic object recognizer algorithms designed to work with imaging sensor inputs require extensive testing before they should be considered robust enough for challenging applications such as military targeting. Testing automatic target recognition (ATR) algorithms in most cases has been limited to a handful of the scenario conditions of interest as represented by imagery collected with a desired imaging sensor. The question naturally arises as to how robust the performance of the ATR is for all scenario conditions of interest, not just for a small set of collected imagery. A way of addressing algorithm robustness is to characterize the input imagery in terms of some common information content or quality measures that can be correlated with ATR performance. This paper addresses the utility of image characterization measures in terms of estimating ATR detection performance by correlation analyses between nine different image measures and the detection responses of two ATR algorithms. Results show that an image measure called target-to-background entropy difference is the best single measure for estimating ATR detection performance, with correlation coefficients as large as 0.60.

Advances in segmentation algorithm design can be guided by the development of uniform methods for segmentation accuracy measurement. A representative set of approaches to segmentation accuracy measurement is presented, and performance is demonstrated on a prototype set of segmented imagery. Unlike the evaluation of the detection and classification phases of the automatic target recognition (ATR) process, segmentation evaluation remains an ill-defined problem in the absence of an absolute, correct segmentation reference. A dynamic assessment of image truth from multiple sources can reduce the uncertainty of the target reference. We describe two approaches to the measurement of segmentation accuracy that make use of internal characteristics of the image in the target reference. These measures begin to demonstrate that the truthing procedure in real-time systems is not necessarily a singlepass reference to a prior model of image truth, but an evolving set of criteria utilizing information from the unprocessed image and from intermediate stages of the segmentation process. Sample measures based on an independent determination of image truth, i.e., hand segmentation of the target regions by the human operator or the manipulation of models of expected target configuration, are also presented. The form of the sampIe measures can be optimized and integrated into a dynamic truthing process to achieve real-time management of the segmentation phase in ATR systems.

This paper presents a simple, nonintrusive technique to separate mechanical jitter and aero-optic jitter from the combined measured jitter. This technique has the advantages of being relatively simple to implement, deriving information from the imaging data, and giving insight into the nature of the flow-field turbulence. The method employs variable collection apertures and a subtraction algorithm designed to separate aero-optic jitter from mechanical jitter. Several levels of vibration and aero-optic effects were measured and then separated in the posttest data processing.

Despite the popularity of digital imaging devices (e.g., CCD array cameras) the problem of accurately characterizing the spatial frequency response of such systems has been largely neglected in the literature. This paper describes a simple method for accurately estimating the optical transfer function of digital image acquisition devices. The method is based on the traditional knife-edge technique but explicitly deals with fundamental sampled system considerations: insufficient and anisotropic sampling. Results for both simulated and real imaging systems demonstrate the accuracy of the method.

The ability of an inexpensive liquid crystal television to modulate both coherent and incoherent infrared light is reported. Experiments demonstrating light modulation for wavelengths between 0.8 μm and 1.1 μm have been performed. Potential applications to dynamic lR scene simulators, compact joint transform correlators, and novel electron trapping processors are described.

One ofthe most critical problems in automatic target recognition (ATR) systems is multiscenario adaptation. Today's ATR systems perform unpredictably, i.e., perform well in certain scenarios and poorly in others. Unless ATR systems can be made adaptable, their utility in battlefield missions remains questionable. We have developed a novel method called knowledge- and model-based algorithm adaptation (KMBAA). KMBAA automatically adapts the ATR parameters as the scenario changes so that ATR can maintain optimum performance. The KMBAA approach has been tested with a nonreal-time ATR simulation system and has demonstrated a significant improvement in detection, false alarm rate reduction, and segmentation accuracy performance.

This paper presents an adaptive or self-learning filter design intended for use in real-time closed-loop pointing control systems engaging multiple targets. The design approach uses a performance index (based on the Mahalanobis generalized distance function) and multiple filters processed in parallel using the same nonlinear measurements as input. Application of performance index criteria to the statistics of individual filter residuals allows the selection of the optimum filter set without the delays typically encountered and thereby allows the composite filter structure to adapt to (or self-learn) uncertainties in modeling target acceleration capabilities. An advantage of this approach is that it also gives an operator (or robotic controller) the confidence level oftracking system performance against a maneuvering target. This information is of interest for deployment of countermeasures (e.g., fire control eventing, alarms, or engagement priority) or simply for laboratory system tests of design adequacy.

Traditional "crossed-grating" moire as well as the newer "sampled- grating" (scanning) moire have proved to be effective methods of shape measurement. There is speculation that the moire patterns of a sampled grating, which are due to aliasing, can be modeled with crossed gratings. We compare the two by writing notationally consistent models of each and show that while crossed gratings can correctly predict the frequencies of a sampled grating, they cannot correctly predict the amplitudes. Our sampled-grating model is a new formulation that accounts for multiple stages of sampling and transmission. We show how neglecting multiple stages can lead to mistakes in moire analysis. We demonstrate our models with an experiment using a digital imaging system.

The optical design space of some simple lenses is investigated systematically. Typical space topographies are visualized with 3-D graphics, where the complete set of available solutions is clearly identified. The space characteristics are then studied and compared through the use of several merit functions with differing degrees of complexity. A two-phase search algorithm, based on global optimization techniques, is proposed here. In the first phase, using a coarse sampling approach, the program finds the favorable regions that correspond to potentially promising configurations. In the second phase, conventional optimization routines are used to find the best solutions in each region. Then an optimum solution is determined according to the application at hand. The proposed algorithm is analyzed and compared to more conventional design approaches. A further refinement of the algorithm excludes from the systematic search some unfavorable configuration regions through the use of a simple expert system. Search times are further reduced through parallel-processing methods. The algorithm provides overall information about a given design space and offers a selection of "best" solutions to choose from. As an example, it is applied to a triplet objective.

The interactions of high-energy laser beams with aerosols are studied in the diffusive regime within the framework of a nonlinear transport equation. A numerical algorithm allows us to advance along the distance of propagation by using spatial steps small compared to the scattering optical depth. We emphasize the effect of laser-assisted imaging, in which—due to the cloud clearing—the roll-off of the modulation transfer function decreases with the distance of propagation. We study the effect of recondensation on the propagation of a low-energy laser beam. The principal findings are the independence of the pulse shape of the turbulence strength and the suppression of the high-energy laser-assisted imaging effect.

This PDF file contains the editorial “Book Rvw: Noninvasive Diagnostic Techniques in Ophthalmology," edited by Barry R. Masters for OE Vol. 30 Issue 02

This PDF file contains the editorial “Book Rvw: Image Analysis Applications," edited by Rangachar Kasturi and Mohan M. Trivedi for OE Vol. 30 Issue 02