The morphological correlation is the optimal method for searching a reference object in an input scene when the figure of merit is based on the mean absolute error (MAE). In practice, it was found that the morphological correlation exhibits high discrimination ability between similar patterns in recognition systems. It is based on threshold slicing the input image as well as the reference filter into many binary slices, as many as the dynamic range of the input permits. The threshold slices of the input and the reference are then correlated and summed up to obtain the morphological correlation. This operation is characterized by a sharp correlation peak but requires many correlation operations. In this work we propose a novel correlation operation that is characterized by even higher discrimination capabilities than exhibited by the conventional morphological correlation and requires less computational effort. The method is based upon the binary representation of the gray level of each pixel in the image. For example: if the dynamic range allows the definition of 256 levels, i.e., 8 binary bits, then a level of 10 will be represented as 00001010. Unlike the morphological correlation, the proposed modification is based on correlating binary slices that are the bitmap representations. Thus, only 8 slices of the input and the reference are required and only 8 correlations rather than 256 performed. The optical implementation of the new approach is fairly simple and can be utilized via the well-known joint transform correlator architecture. Experimental results demonstrate the advantages of the suggested method.

Nonlinear joint transform correlators (JTCs) have been proposed for optical information processing. They have been shown to be attractive in many difficult instances because of their high discriminating performance. However, unlike the linear matched filter, which was designed on the basis of the statistical estimation theory before its implementation in optical correlators, investigations on nonlinear filtering techniques have been mostly experimental and their basic properties in terms of signal processing and pattern recognition still need theoretical analyses. We propose in this paper to analyze the optimal solutions obtained in the context of the statistical estimation theory when the spectral density of the additive Gaussian noise is unknown. Maximum likelihood, maximum a posteriori and Bayesian solutions to this problem are discussed and practical consequences are analyzed. In particular, we show that nonlinear JTC methods can be considered as a first order, but very efficient, approximation of these optimal solutions.

In image recognition applications, complex decision regions in the image space are needed. Linear filtering forms the decision regions by hyperplanes in the image space. We determine the decision region formed by Fourier plane nonlinear filtering. In the case that power law nonlinearity is applied in the Fourier plane, the decision region turns out to be approximately an n-dimensional parabola which opens toward the direction of the reference vector. That is, the intersection of the decision region with any plan (two dimensional vector space) not containing any vector parallel to the reference vector, is a bounded convex region enclosed by a closed curve. The size of the convex region depends on the filter nonlinearity, which determines the distortion robustness and discrimination capability of the filter. It can be adjusted by choosing different Fourier plane nonlinearities and/or different threshold values at the output plane. These types of regions are desirable and well suited in image recognition.

In this paper we design an optimum receiver to detect a pattern or a reference signal. We design a receiver which detects the signal distorted by a multiplicative noise on the signal itself, as well as by additive noise and by non- overlapping scene noise. We use hypothesis testing to determine the location of a target buried in multiplicative, additive, and non-overlapping background noise. We also consider the case when the reference pattern is illuminated with an unknown illumination constant. Using computer simulations we show that, for the images tested here, the optimum receiver performs better than some of the existing receivers.

The theory of correlation filters does not make any assumptions about the sensor or image format. Thus the same class of algorithms is readily applicable to multiple sensor environments such as IR, SAR, LADAR, or CCD (visual). The advantage is that the same theory is valid for multiple sensor applications, the processing steps are common (and code) are re-usable in different sensor platforms, and the algorithms are rapidly re-trainable. The paper points out the key benefits resulting from the general formulation and solution resulting from the correlation approach to ATR.

Recently, optimal algorithms for locating a target on nonoverlapping background, based on maximum likelihood approach, have been designed. In particular, different ways of modeling the target have been proposed. When the gray levels of the target are known, the reference of the target can be modeled as a deterministic function. On the other hand, when the gray levels of the target in the input image are unknown or can vary from one image to another one, the reference of the target has to be considered as a pattern with random gray levels. Moreover, it is possible to unify both the deterministic and the random target approaches into a single model, where the target is modeled using a linear combination of deterministic values and random variables. Based on this model, we propose to design an algorithm that optimizes the likelihood ratio between the two hypothesis that a target is present and that it is absent within a small sub-window of the image. We show that this technique is more efficient than the maximum likelihood approach when the noise statistic of the background is strongly nonhomogeneous, which is the case in many real-world images. The presented algorithm is based on correlations and it can thus be implemented in an architecture using optical correlators.

A rotation invariant binary circular filter has been developed for optical pattern recognition. The filter is generated using an iterative numerical optimization method. The optimization is based on the genetic algorithm, which fits very well in optical systems due to its parallel nature. The features of the genetic algorithm provide a highly efficient and rapid learning process. During training, the parameters of a binary circular filter are selected to maximize the distinction between the target and other expected objects in the image. The genetic algorithm is searching through the complete filter space for the global solution, this is the filter with the best performance. These iteratively designed filters are good discriminators because they utilize all the spatial visual information about the target. The design of the rotation invariant filter does not require any a priori information about the target image. The rotation invariant filters are designed as binary circular filters to be suitable for real- time applications, when combined with spatial light modulators.

The presence of turbulence such as maritime aerosols between the target and the observer degrades the detection and classification performance of electro-optical sensors and detectors. A filtering algorithm that takes into account of the environmental degradation, the background non-overlapping noise, the non-stationarity of the scene, non-target objects and additive system noise is designed and implemented to detect and classify targets under such conditions. The detection performance of this algorithm has been validated using computer simulations and found to be superior to filters that are optimal with respect to noise statistics but do not take into account the effects of environmental conditions. The algorithm is then extended using a set of training images to be distortion-invariant with respect to different target aspects.

The design of filters for pattern recognition that have optimal trade-offs among the criteria of noise robustness, sharpness of the correlation peak, and Horner efficiency when input scene noise is spatially disjoint (nonoverlapping) with the target are presented. Computer simulation is made to illustrate filter performances for optical pattern recognition.

Intelligent Debris Analysis (IDA) requires significant time and resources due to the large number of images to be processed. To address this problem, we propose a hybrid optoelectronic and computer vision approach. Two major steps are involved for IDA: patch-level analysis and particle level analysis. An optoelectronic detection system using two ferroelectric liquid crystal spatial light modulators is designed and constructed to perform path-level analysis, and advanced computer vision techniques are developed to carry out more intelligent particle-level analysis. Since typically only a small portion of the debris filters require more sophisticated particle-level analysis, the proposed approach enables high-speed automated analysis of debris filters due to the inherent parallelism provided by the optoelectronic system.

This paper is concerned with a method of non-contacting measurement of mechanical strain within specimen. It describes a new optical set-up to perform high speed digital laser- speckle correlation with the ultimate aim to deduce surface element displacements associated with the translation of laser-speckles emanating from those surface elements. The novel optical set-up combined with the application of line scan cameras attached to digital signal- or very fast general- purpose processors allows measurement rates that for most practical purposes are only limited by the integration time of the camera necessary to obtain properly exposed images. Instead of obtaining a two-dimensional vector by searching for the best space-lag for a digitally calculated cross- correlation estimate of the initial and translated speckle images a single component of that vector parallel to the straining direction is obtained by finding the space-lag of optically preprocessed almost one-dimensional speckle fields. The necessary optical preprocessing is performed in the Fourier-plane of the imaging optics. This way the numerical complexity of the algorithm is greatly reduced resulting in lower processing time per frame. System considerations for practical strain measurements are detailed and the measured sensitivities are presented.

We explain that a holographic filter of the generalized Hough transform can be easily obtained by the use of multiplexing techniques in hologram recording. To show the feasibility of our approach experimentally, we recorded the Hough transform filter of both line and circle parameterizations by combined use of rotational and angle multiplexing. Experimental results on the Hough transform for a few input patterns are presented.

A high speed optical correlator is presented in this paper. It is a joint transform correlator using a BSO photorefractive crystal in the Fourier plane. The performance of the system such a rotation and scale robustness are presented for fingerprint recognition. To demonstrate the interest of such an optical processor, a comparison with numerical systems is presented. Besides, we will also show that the evolution of correlators is quite compatible with the evolution of numerical processors.

We review GaAs Fabry-Perot vertical cavity electro-optical devices constructed at Lockheed Martin Sanders, with particular attention to spatial light modulators and discuss a number of their applications. Our manufacturing processes enable the construction of large format, highly-uniform optical modulator and smart pixel arrays integrated with silicon CMOS VLSI circuitry. These devices can be used for a wide variety of applications including: optical computing for target recognition and signal processing, optical data routing, optical interconnect systems, and optical memory storage and access.

Barriers to commercialization of optical correlators include the complexity and cost of their manufacture, their large size compared to typical electronic processors, and the cost of their components. Using sub-micron CMOS VLSI fabrication processes it is possible to build the two SLMs and photodetector array of a Vander Lugt correlator on a single silicon die. The correlator's lenses can be fabricated on a single piece of glass using diffractive optics technology and then attached to the CMOS die to form a monolithic assembly. This approach greatly reduces the mechanical degrees of freedom that must be controlled by the correlator's housing thus lowering cost, reducing size, and improving reliability. Here we report on the design and performance of a prototype.

An optical photorefractive joint transform correlator (PRJTC) was built using a twisted liquid crystal spatial light modulator in the input plane to display the images and a photorefractive crystal in the Fourier plane to perform the nonlinear correlation. We present here new correlation filters to optimize the correlation. There filters are correlated with the scene instead of the simple reference. To calculated these filters, we introduce two characteristics of the setup to optimize the filters: the nonlinearity of the photorefractive crystal, the coding domain of the displaying device.

Smart-pixel spatial light modulators (SLMs) with optical inputs and outputs can offer a powerful combination of optical and electronic processing functions. Here we report on work in progress on an experimental FLC-VLSI SLM intended for use in the first Fourier plane of a joint transform correlator (JTC). The JTC SLM is designed to eliminate the unwanted autocorrelation term that normally exists in the JTC's output. One of the JTC's input images is modulated in phase or amplitude causing a modulation term to occur in the intensity of their joint transform. The smart-pixel SLM selectivley responds to the modulation term and displays the result as an image that will pass through the second and final Fourier transform lens (each pixel contains both a photodetector and an optical modulator).

We demonstrate a novel single shot autocorrelation technique for characterization of ultrashort pulses. Unlike existing single shot autocorrelation techniques, our new technique is capable of characterizing optical pulses over a femtosecond to picosecond pulse-width range. Our technique uses a grating at the entrance of the system, introducing a Transverse-Time- Delay (TTD) into the reference pulse. The pulse front in the resulting field is decoupled from the wave front. The signal pulse to be characterized and the TTD reference pulse are mixed in a nonlinear optical crystal, producing a second harmonic field whose transverse spatial extent is proportional to the signal pulse width. Since our technique allows for decoupling of the time delay from the propagation direction (unlike the commercial single shot autocorrelators), we can select the angle between the intersecting pulses to satisfy the phase matching conditions, achieving best efficiency while setting the resolution independently in the orthogonal direction. In addition, by controlling the slope of the TTD, the system can adapt to a wide range of input pulse widths. In this paper we will present the basic principles as well as experimental results for this new autocorrelation technique.

Transformation paths (TPs) allow to transform 1-D shuffle interstage interconnections into their topologically equivalent 2-D shuffles and vice versa. By the assumption of different TPs in source arrays (SAs) and destination arrays (DAs), respectively, TPs exist through switching arrays independent of form (square or rectangle) and size. The ongoing work is devoted to scalable algorithms for arbitrary large photonics systems.

A novel all-optical parallel switching device was proposed based on photoinduced changes of a complex refractive index in a guided mode thin film composed of silver and polymer containing organic dyes. An incident angle of a probe beam was set at a value corresponding to the minimum reflection due to a guided wave mode. Pulsed laser excitation caused changes of absorption of the polymer films due to excited state formation or photochromism, which resulted in the increase of reflection because the guided mode condition was changed. According to the Kramers-Kronig relationship the real part also changes, which results in the shift of the minimum. We can select the wavelength for all-optical switching in this geometry. We used phthalocyanines and photochromic spiropyran dispersed in poly(vinyl alcohol) or polystyrene. For the former system, transient absorption due to the excited triplet gave a highly reversible very fast modulation of green light upon repeated excitation by ns laser. For the latter, self-held switching was achieved upon UV or visible laser excitation with response times less than 20 ns. The details of these systems will be presented.

Massive parallel optical interconnection systems in the 3-D physical space offer advantages but they also represent serious difficulties caused by the increase of the degree of freedom. These difficulties are mainly (1) packaging and precise mounting but also (2) problems of combinatorial nature. Throughout the paper, the presentation of routing information in greater than or equal to 2-D MINs is discussed and the results are aimed to avoid delay and additional hardware. The results apply to distributed switching and processing under the optical interconnection regime.

We describe an optical learning three-layer neural network that uses a back-propagation algorithm for human face recognition. Learning is performed by two-dimensional optical means for handling images without scanning and pixeling, because a face recognition problem needs a large-scale network. The two-dimensional input image is treated directly without vectorizing so that a recognition process of an unknown facial image is executed in real-time. We demonstrate training and recognition of three human faces using an experimental system.

The phase-encoded joint transform correlator as a decoder of an optical double random phase encoding system is described. By using the phase-encoded joint transform correlator, it is possible to decode the encrypted image with the same random phase mask which was used in the encryption procedure. The decoding algorithm and optical systems are described. Computer simulations confirm the proposed decoding algorithm.

Fourier optical pattern recognition has wonderful properties (high speed, space-invariant operation, low power consumption, target location) and one terrible property (It can work perfectly only for the rare, uninteresting case of two linearly separable categories). More powerful discrimination methods lack the other wonderful properties. I show here how to have it both ways at once. By using roughly the Vapnik- Chervonenkis (VC) dimension number of properly trained Fourier filters in parallel and performing pixel-by-pixel thresholding on the output planes, we can assemble a net output plane which achieves provably optimum, predictable discrimination on any sets.

A stereo vision system, as it receives two images equivalent to the one shown on left and right human eyes, can provide 3-D effect. However, the stereo disparity caused by the different parallax of the two images makes an observer feel fatigued and reduces the 3-D effect. Therefore, this paper presents a new approach to keep the stereo disparity to be zero via a JTC- based adaptive tracking of a moving object. In this method, the optical JTC system tracks the relative locations of a moving objects via measuring the correlation peak of the two images. Through some optical experiments the proposed stereo vision system is proved to be insensitive to background noises and control the convergence angle in real-time.

The security verification method using a transform random phase mask as an optical mark bonded to a document or other product is proposed. This mask consists of separated and shifted fragments of a reference phase mask. If the transformed and the reference masks are entered into an optical correlator, the autocorrelation peaks series is produced on the correlator output. The distances between peaks and the peak intensities were used to produce the feature vector. Identification of the document or other product takes place if the feature vector and the reference feature vector coincide. The procedure of the transformed mask generation and the process of the peaks' producing in a conventional joint transform correlator were considered. The advantages of transformed mask applications in optical correlators are discussed. The joint transform correlator experimental setup containing the spatial light modulator PRIZ was designed and the optimal conditions to produce the autocorrelation peaks were found.

Among a variety of optical correlators, the joint transform correlator (JTC) has some advantages over others and has been extensively studied. However, in some cases, a direct correlation of the input image with a reference image would not be able to find the matched pattern. Recently, morphological hit-or-miss transformation (HMT) has been used to improve the pattern recognition ability of JTC. In this paper we use the concepts of uncertain pixel to improve the distortion-invariant ability of morphological hit-or-miss transformation used in JTC. The novel method is called joint rank-order multi-value hit-or-miss transformation (JRMHMT), which reduces the decision effect of the pixels that is easily disturbed by rotation and scaling distortion. Without losing information of the input images, JRMHMT can realize the precision recognition between two images and has a better distortion-invariant ability in scaling and rotation than the ordinary HMT methods. The comparisons of JRMHMT with the ordinary HMT methods in theoretic analysis and experiments are given in this paper. Based on a single lens JTC and a novel multi-value complementary encoding method, joint rank-order multi-value hit-or-miss transformation correlator (JRMHMTC) is constructed for realized MVHMT in one step. The correlated result image can be thresholded with a high-level threshold so that correct justification can be achieved. The simulation results and experimental demonstrations are both listed.

A novel compact BNI architecture in free space is composed of symmetrical arrays of binary phase grating having advantages of high spatial density, no coupling material, wide bandwidth, low insertion loss, low crosstalk, compacting system, easy cascading, etc.

The artificial neural network (ANN) technique was successfully used to make correlation signal post-processing of the real- time 3-D object intra-class distortion-invariant correlation recognition system. The Synthetic Discriminant Function (SDF) filtering technique was used to perform the distortion- invariant correlation. The real-time Spatial Light Modulator (SLM) was used as the scene input device and filter sequencer in the real-time optical correlator. The correlation signal post-processing results show that this technique was very effective to extract the signal from noise background of the correlation plane. In compared with the conventional recognition methods, the intra-class recognition possibility and reliability were improved greatly.

We present a study on a high-speed optoelectronic system for implementing space variant transforms (SVT) in image and signal processing using a Hough Transform (HT) as an example. The HT has been found to be highly useful in applications requiring detection of lines, ellipses and hyperbolic shapes, such as radar detection and data fusion, topographical map analysis, etc. However, the implementation of a SVT such as HT, is computation and memory intensive, e.g. HT of an image of dimension N X N requires greater than N³ operations. All-electronic systems remain inadequate when real time SVT processing of large data sets is required. In this paper we show that an optoelectronic (OE) system employing parallel processing can perform such SVT requiring on the order of only N steps. We show that our proposed OE system can HT an input image of dimension N equals 1024 in 2.1 ms.

Composite classifiers that are constructed by combining a number of component classifiers have been designed and evaluated on the problem of automatic target recognition (ATR) using forward-looking infrared (FLIR) imagery. Two existing classifiers, one based on learning vector quantization and the other on modular neural networks, are used as the building blocks for our composite classifiers. A number of classifier fusion algorithms are analyzed. These algorithms combine the outputs of all the component classifiers and classifier selection algorithms, which use a cascade architecture that relies on a subset of the component classifiers. Each composite classifier is implemented and tested on a large data set of real FLIR images. The performances of the proposed composite classifiers are compared based on their classification ability and computational complexity. It is demonstrated that the composite classifier based on a cascade architecture greatly reduces computational complexity with a statistically insignificant decrease in performance in comparison to standard classifier fusion algorithms.

Emblems using holograms or other diffractive devices have long been used to mark cards and other objects as a means of authentication. The effectiveness of such emblems as a security device is ultimately determined by the inspection system. Due to the expense and highly variable performance of the human inspector, automated machine reading devices are an attractive alternative for performing the verification task. An additional advantage of the machine reader is that information regarding the card or its holder may be stored covertly. A security verification system is presented consisting of a holographic security emblem in which information is covertly stored, and an automated reader based on a joint transform correlator (JTC). A holographic encoding method is used to produce an emblem that stores the required phase and/or amplitude information in the form of a complex, 3-D diffraction pattern that can only be interpreted through the use of a second 'key' hologram. The reader incorporates the use of the non-linear material, Bacteriorhodopsin, as a means of miniaturizing the system, reducing system cost, and improving system performance. Experimental results are presented that demonstrate the feasibility of the approach for security applications.