Pattern recognition of ISAR small ship images is considered. These represent a formidable new distortion-invariant pattern recognition problem. Variations to the standard image formation steps were used. The preprocessing used is briefly noted. These include new techniques to produce data with horizontal deckline and proper upward superstructure. New algorithms to determine if an input image is useful are developed. These were necessary for such data. New variations of standard distortion-invariant filters are developed to provide excellent recognition results.

Jet Propulsion Laboratory has been developing grayscale optical correlator (GOC) for a variety of automatic target recognition (ATR) applications. As reported in previous papers, a 128 X 128 camcorder-sized GOC has been demonstrated for real-time field ATR demos. In this paper, we will report the recent development of a prototype 512 X 512 GOC utilizing a new miniature ferroelectric liquid crystal spatial light modulator with a 7-micrometers pixel pitch. Experimental demonstration of ATR applications using this new GOC will be presented. The potential of developing a matchbox-sized GOC will also be discussed. A new application of synthesizing new complex-valued correlation filters using this real-axis 512 X 512 SLM will also be included.

A technique for the experimental implementation of fully complex filters with commercially available spatial light modulators (SLMs) is reported. The filters are incorporated into an all-optical correlator and a hybrid digital-optical correlator, the relative merits of each configuration being considered. Various filter functions requiring complex modulation are demonstrated, consideration being given to the degradation of filter performance due to the limited quantization and dynamic range with which they can be implemented using current SLM technology.

We show how the signal to noise ratio distributes ideally in the complex plane of filter values, and we show how it is captured in its representation on the restricted set of values the filter SLM is able to realize. The ability to take strong advantage of a large dynamic range of filter magnitude is apparent. Further work will extend this concept to other metrics of optical correlator performance, including statistical pattern recognition criterion functions such as Bayes' error, ROC (receiver operating characteristic) curve's area, and Fisher ratio.

We describe several optoelectronic methods based in digital holography for recognition of 3D images. The phase and amplitude of a Fresnel diffraction pattern of a 3D reference object is measured with digital holography. This complex information is compared with that coming from a similar digital hologram of a 3D input scene using correlation techniques. In this way, the method allows us to detect the presence of the 3D reference object in the 3D input scene with high discrimination. Pattern recognition techniques that are shift-variant or shift-invariant along the optical axis are described. In the later case it is possible to detect the 3D position of the reference in the input scene with high accuracy. Using the information contained in the digital holograms it is also possible to measure small 3D orientation changes of a 3D object. Experimental results are presented.

We present a high-speed compact Binary Phase Joint Transform Correlator system based on a single liquid crystal over silicon spatial light modulator. The system is capable of processing images of 320*120 pixel resolution at frame rates currently limited to around 40 frames per second by the choice of camera within the system. The system is presented in the context of an image comparator system in a fighter aircraft cockpit, which is used to track the view of the pilot. This is achieved by using a helmet-mounted camera to provide the input scenes and some of the inherent properties of the Joint Transform Correlator. Results from an experimental prototype are presented.

JPL is developing a high-density, nonvolatile Compact Holographic Data Storage (CHDS) system to enable large- capacity, high-speed, low power consumption, and read/write of data for commercial and space applications. This CHDS system consists of laser diodes, photorefractive crystal, spatial light modulator, photodetector array, and I/O electronic interface. In operation, pages of information would be recorded and retrieved with random access and high- speed. In this paper, recent technology progress in developing this CHDS at JPL will be presented. The recent applications of the CHDS to optical pattern recognition, as a high-density, high transfer rate memory bank will also be discussed.

If an optical correlator is to perform at full potential, the filtersmith must know what complex action will result from the control he applies to the filter SLM. If the SLM is spatially variant (and all are, to some degree or other), the behavior may be different at every frequency plane pixel. We have previously reported characterization of the full-complex behavior at every pixel of the SLM. We have refined the method in two distinct ways: we are doing multi- step interferometry (rather than only phase quadrature), and we have significantly improved the isolation of an individual pixel's complex action.

In a previous report we developed optimization algorithms showing how optical correlation filters operating with obscured inputs were affected by disjoint constant background illumination. In this paper we extend these studies by upgrading our algorithms to include the theoretical treatment of zero-mean disjoint noise, as well as constant background illumination. Representative cases of computer simulations involving either noise clutter or background illumination are used to characterize the performance of our upgraded algorithms.

This paper presents an incoherent optoelectronic processor which is able to segment an object in a real image. The process, based on active contours (snakes), consists in correlating adaptive binary references with the scene image or with a preprocessed version of the scene image. The proposed optical implementation of algorithms which are already operational numerically opens attractive perspectives as far as speed is concerned. Furthermore, this experiment is a new application for optical processors.

In this paper, we present an algorithm for fast calculation of the normalized cross correlation and its application to the problem of template matching. Given a template t, whose position is to be determined in an image f, the basic idea of the algorithm is to represent the template, for which the normalized cross correlation is calculated, as a sum of rectangular basis functions. Then the correlation is calculated for each basis function instead of the whole template. The result of the correlation of the template t and the image f is obtained as the weighted sum of the correlation functions of the basis functions. Depending on the approximation, the algorithm can by far outperform Fourier-transform based implementations of the normalized cross correlation algorithm and it is especially suited to problems, where many different templates are to be found in the same image f.

This paper describes a method of noise elimination in a joint transform correlator (JTC). In the JTC the final output the correlation peaks, are immersed in the noise hence often it is difficult to detect. In this technique final output is grabbed on a CCD camera and is programmed to perform change detection between real time correlation output and a static output of background noise. The change detection between these two frames completely eliminates the static and DC noise. Electrically addressed spatial light modulators having 640 X 480 resolution elements, are used in object and Fourier planes. Laser diode of 5 mW is used for modulation of images and illumination of Joint Power Spectrum. The results are captured with the help of CCD cameras and displayed on monitors. Complete noise elimination has been achieved in a final output by this method and the noise free correlation peaks are obtained. The best result achieved by digitally processed correlation peaks and choosing an optimum intensity value to suppress the noise and enhancement of the correlation peaks.

We show that certain additions to the mathematical model of a correlator can improve the signal to noise ratio of the optical correlator itself. All computational optimizations include a digital simulation process as a critical tool. The more accurate the model of the optical correlator, the better the optimization. improvements to our correlator model, based on correlator measurements, are made to the simulator in our filter optimization process. Comparisons of the signal to noise ratio are made both digitally and optically. Laboratory results are given.

In this paper we theoretically analyze and demonstrate that a photorefractive correlator, originally proposed by D. M. Pepper and later implemented by J. O. White and A. Yariv as well as many others, can be used to realize adaptively a wide variety of optimal correlation filters such as the matched filter, the inverse correlation filter, the maximum discrimination correlation filter and several trade-off correlation filters.

In this paper we introduce the mean-square-error correlation filter for obscured targets. We derive from this algorithm that the DC-blocked phase-only filter is the best practical approximation to this filter for recognizing SAR images with enhanced scattering centers for both inputs and templates. The performance of the DC-blocked phase-only filter is tested in the recognition of SAR images from the MSTAR data base under extended and non-extended operating conditions. The heterogeneous correlation filter algorithm used to correlate totally different images is utilized to correlate distorted images under extending operating conditions.

We address the problem of small target detection in active polarimetric images. We consider polarimetric imagers which illuminate the scene with purely polarized light and form two images with the part of the backscattered light polarized in the same state as the illumination, and in the orthogonal state. We first show that if the illumination intensity is considered random, the only information relevant for target detection is contained in the ratio of both images. Since the detection algorithm optimal in the maximum-likelihood (ML) sense is computationally complex, we propose a simpler yet efficient target detection algorithm, which consists in applying the ML detector optimal for additive Gaussian noise to the logarithm of the channel ratio. Indeed, in this new image, the fluctuations can be considered additive and their distribution nearly Gaussian. We show that this method performs well in both simulated and real polarimetric images.

Target recognition is an important task for many automatic systems based on imagery. The recent technique of active contours (snakes) is well adapted to the segmentation step when the recognition is made from the shape of the target. Classical segmentation strategies are generally edge-based in the sense that the segmentation is driven from an edge map of the scene. Consequently, these methods which are efficient with a certain class of problem could fail in presence of strong noise. We have recently proposed an original approach for the statistical segmentation of an object (statistical snake) for which the image is assumed to be made of two regions (the object and the background) composed of homogeneous intensity random fields. In this article, we characterize the quality of the segmentation as a function of the target resolution and noise level with two similarity measurements based on Hausdorff distance between the exact contour and the result of the segmentation.

In this paper, we propose an optoelectronic fringe-adjusted joint transform correlator (JTC) based two-layer neural network for invariant face recognition while accommodating in-plane and out-of-plane 3D distortions. The neural network is utilized in the training stage for a sequence of facial images and through a process of supervised learning in order to create composite images that are invariant to 3D distortions. The proposed technique is implemented by using the FJTC technique. The FJTC technique has been chosen due to its superior performance over alternate JTCs and the feasibility of its implementation in the all-optical domain. The simulation results obtained from the proposed technique are then compared with those obtained using alternate techniques (such as using synthetic discriminant functions). The fringe-adjusted JTC based neural network technique has been found to be more efficient and yields better results than the synthetic discriminant function based technique.

Shape parameters based on circularity, Fourier descriptors, and invariant moments are studied for the automatic detection of weapons in millimeter-wave data. The data is collected by a 30-frames-per-second millimeter-wave (MMW) imager manufactured by Trex Enterprises for the detection of weapons concealed underneath a person's clothing. Results are illustrated through processing real MMW data.

A metric is developed for evaluating performance degradation of edge detection algorithms as a function of signal to noise ratio (SNR). The metric combines both missed detections and false alarms to form a composite score. This provides a basis for objectively comparing the performance of different techniques and quantifies relative noise tolerance. It is applied to various popular algorithms, Sobel, Roberts, Prewitt, and Laplacian of Gaussian, but is described in sufficient detail to facilitate easy application to other edge detection methods. Results shown allow selection of the most optimum method for application to images with known SNR levels.

The theory of designing the optical-electronic image processing computer systems has been presented. A model of parallel image processing system has been considered, that is based on the principle of function decomposition. The implementation possibilities of different image processing operations with the help of optical and electronic computer means have been analyzed. A structure model of computer system has been examined, that is a conveyor of parallel computer devices. The evaluation of time outlay in the system, while processing an image or a series of them has been made. The differences of time outlay from conveyor length change and the correlation of optical and electronic devices and processing time in them have been exposed. The designing method of image processing systems in static made has been elaborated. There are presented the results of investigations of the influence of the median square deviation, the influence of time of the processing in the modules on the throughput capacity of the system under the different electronic and optical modules quantity. According to the results of investigations the recommendations of increasing the system's throughput the capacity are formulated. On the basis of these recommendations, the system design method of image processing in the dynamic mode is elaborated.

In this article, we propose to return to the simplest optical architecture and to transfer all the recognition intelligence in to the filter design. However the use of such a filter to carry out multiple correlations in parallel poses the problem of product space bandwidth limitation in the filter plane (Fourier plane): information on references is injected into the filter by a frequentially multiplexing of the references constituting the training base. From this point of view, we propose an optimization of the segmented filter in order to increase its performance and its capabilities.

A novel approach is proposed to improve the discrimination ability of the radial harmonic filter for shift- and scale- invariant pattern recognition. The approach combines the morphological correlation and the radial harmonic expansion for shift- and scale-invariant pattern recognition with improved discrimination. In the approach, the reference gray-level image is first decomposed into a series of binary images, and then the radial harmonic expansion is applied to these binary images. The morphological radial harmonic correlation is achieved by correlating the single-order radial harmonic components with the corresponding sliced binary images obtained from the input scene and summing the correlations. Simulation results show that the MRHC is shift and scale invariant and the discrimination capability is improved compared to the normal radial harmonic correlation.

A spatial light modulator can be used to actively illuminate moving objects with spot patterns. This research studies the spot patterns designed to detect and track object motion. To enhance the detection and tracking process, a system supervisor dynamically updates each successive spot pattern. The spot patterns are implemented by pseudo-randomly encoding their frequency plane into a phase-only filter to be implemented on a 128 X 128 Boulder Nonlinear System's LCD device. The modification of the spot patterns are based on the change of reflected intensity from one frame to the next. Detection, tracking, encoding and supervisor design and methodology are presented.

Visual cryptography made it possible to decrypt the information encrypted by thresholding scheme not with digital system but with human vision system. This method, however, has some limit in it because of the rack of resolution in both the spatial and amplitude domain. Optical visual cryptography, which used laser system instead of human eyesight, was proposed by conjunction of the optical theory with the cryptography. However, it also had some difficulties because it did not overcome the existing problem of visual cryptography completely. The problems occurred in the process of transferring data processing system from visual to optics. Therefore, it is appropriate to approach these problems in terms of optics. The results show that the optical visual cryptograph system has both the effectiveness and reliability as well as real-time implementation property.

Historic information regarding the appearance and creation of fundamentals of algebra-logical apparatus-`equivalental algebra' for description of neuro-nets paradigms and algorithms is considered which is unification of theory of neuron nets (NN), linear algebra and the most generalized neuro-biology extended for matrix case. A survey is given of `equivalental models' of neuron nets and associative memory is suggested new, modified matrix-tenzor neurological equivalental models (MTNLEMS) are offered with double adaptive-equivalental weighing (DAEW) for spatial-non- invariant recognition (SNIR) and space-invariant recognition (SIR) of 2D images (patterns). It is shown, that MTNLEMS DAEW are the most generalized, they can describe the processes in NN both within the frames of known paradigms and within new `equivalental' paradigm of non-interaction type, and the computing process in NN under using the offered MTNLEMs DAEW is reduced to two-step and multi-step algorithms and step-by-step matrix-tenzor procedures (for SNIR) and procedures of defining of space-dependent equivalental functions from two images (for SIR).

Edges intersect in an edge image and this intersection point may correspond to 3D point joining two straight lines in the real world scene and those lines represent a real object plane; in this case it is called a real lines intersection, otherwise it is called a virtual intersection. An automatic system for locating image lines is likely to produce many virtual intersections and so despite many studies in the field of boundary recognition, the question of whether the intersection of two liens in an image of a 3D scene corresponds to a real object point still merits further investigation. This paper presents a computational technique to identify the real or virtual nature of the edge intersections. The discrimination is based on rectified images obtained from a pair of uncalibrated images. The method is tested using different types of images of real scenes. The results obtained showed reliable decisions.