One of the major barriers to commercial application of optical technology to information processing is the high cost of system development and manufacture. This problem has been solved in other industries through the use of computer aided design (CAD) and integration of system design with manufacturing. The development of better system level metrology is needed to allow more computer-based methods to be used in this process. As a test case, we are designing an optical pattern recognition system to be performed on an input image (at video rates) versus a large reference set, for example 1000 faces, with images of 640 by 480 pixels or larger. We have constructed both an optical pattern recognition system and a holographic memory system which we have instrumented and used to address the metrological needs of these applications. This has allowed us to evaluate the level of system and component level metrology needed for real-time video processing. This report addressed the metrological issues encountered in building and testing these systems.

This paper reviews a number of optical information processing techniques for encryption, security and anticounterfeiting.

Biometric Encryption^TM is an algorithm which has been developed to securely link and retrieve a digital key using the interaction of a biometric image, such as a fingerprint, with a secure block of data, known as a Bioscrypt^TM. The key can be used, for example, as an encryption/decryption key. The Bioscrypt comprises a stored filter function, produced by a correlation-based image processing algorithm, as well as other information which is required to first retrieve, and then verify the validity of, the key. The process of securely linking a key with a biometric is known as enrollment, while the process of retrieving this key is known as verification. This paper presents details of the enrollment and verification procedures.

Many applications, including military, medical, and security, have a requirement for small, low-power, low-cost pattern recognition systems that are capable of locating and identifying targets or anomalies. Optical correlators can perform two-dimensional pattern recognition at greater rates than digital platforms of equivalent size, power and/or weight. The patented Miniature Ruggedized Optical Correlator (MROC^TM) has been built to meet the environmental, size, power, and weight requirements of rugged military and commercial applications, and at a cost that will permit wide deployment of the capability. The third generation of the optical correlator (MROC^TM III) includes a ferro-electric liquid crystal (FLC) spatial light modulator (SLM) device in both the input plane and the filter plane, and an improved block design that results in both high light efficiency and very fast operational speed. This new correlator design has a reduced volume of approximately 260 cubic centimeters. The MROC III module, which includes all drive electronics and interfaces developed for the MROC II, is a 6U VME module that now occupies only four VME card slots compared to the 5 slots required for MROC I and MROC II. In this paper we will provide a brief review of the MROC architecture and present sample results for the MROC III. Our initial tests demonstrated very high correlation levels, i.e. excellent discrimination (SNR approximately equals 50), at pattern matching rates of 1920 correlations per second (cps).

A real-time optical joint transform correlator (OJTC) has been developed for direct-input optical correlation operation. Innovative features of this OJTC include: (1) two separate spatial light modulators, one for large-frame video rate input, the other for high-speed reference updating; and (2) a multiple quantum well holographic device (MQHD) for high-speed joint transform power spectrum detection and read-out. These system innovations have enabled real-time adaptive optical joint transform correlation to accommodate scale, rotation, and other variations required for real-world pattern recognition applications. System architecture and analysis will be depicted. Experimental results of invariant landmark tracking using this OJTC will be provided.

For optical processor applications outside of laboratory experiments, it is desirable to streamline the data flow in order to obtain the highest possible throughput from the system. This paper presents the data flow architectures for two optical processors designed and built by Boulder Nonlinear Systems, as well as the processor designs and some experimental data.

A compact (Camcorder size), ultrahigh-speed, grayscale optical correlator capable of using grayscale input and real-valued correlation filter is reported. The unique bipolar-amplitude correlation filter modulation capability has enabled the direct implementation of composite filter algorithm such as MACH[8]. The direct usage of grayscale input also eliminated the need of image preprocessing (e.g. Sobel filtering used for binary optical correlator). System architecture of this grayscale optical correlator will be described. Experimental results obtained using optically implemented MACH filter, for the first time, will also be provided.

We present a new means of creating a binary phase only 1/f joint transform correlator using a ferroelectric liquid crystal spatial light modulator. By using the two-pass technique of the 1/f JTC in combination with a binary phase FLM SLM it is possible to build a mechanically robust correlator with a correlation rate in excess of 2 kHz. By using a checkerboard phase encoded input, it is possible to use the FLC SLM as a binary phase modulator while also avoiding the problem of CCD camera saturation due to zero order. An edge enhancement algorithm is then used on the captured image to produce a suitable image for the second pass through the correlator. The algorithm has been chosen to produce a near perfect binary phase image which results in a correlation plane with greatly suppressed zero order (or dc terms).

A method is presented for selecting the optimum polarizer configurations for a complex optical correlator. Two twisted nematic (TN) liquid crystal televisions (LCTVs) are used to modulate intensity and phase. The profiles of intensity and phase modulation vs. applied pixel voltage are determined using a Jones calculus model of the Philips P3.4 TN LCTV between polarizers. This allows selection of separate polarizer orientations which yield the highest ranges of intensity and phase modulation. The optimum configurations provide continuous intensity modulation with approximately 261 degrees of phase modulation.

Two specialized silicon photo-diode detector arrays and supporting electronics have been developed for use in the Fourier plane of a coherent optical image processor. Each detector array has geometrical characteristics tailored for extraction of specific image features. The first detector performs angular measurement of in-plane rotation of simple objects to an accuracy of better than one degree. This detector is being integrated into a system to measure the attitude of missiles in flight. The second detector uses the well-known ring-wedge geometry to measure edge angles and power spectra in the Fourier plane. This detector and supporting neural network software forms a powerful automatic target recognition system. Both detectors are capable of processing images at rates exceeding one thousand frames per second. We describe systems that process NTSC or VGA video images at 60 frames per second.

This paper describes a method to modify the point spread characteristics of an imaging system to perform convolution filtering of incoherent image fields prior to detection. The technique utilizes an aperture plane phase only optical element (kinoform) which is computer generated to optimize efficiency and is fabricated as a bleached hologram. In addition to providing a high speed alternative to digital enhancement of images, optical processing using this approach has several interesting properties. The most significant of these is the ability of the phase element to retain and process high spatial frequency image information from parts of the image which would otherwise be out of focus. As a result this technique allows an optical implementation of three- dimensional convolution filtering, a practical demonstration of which is given in this paper.

Distortion-invariant correlation filters are used to detect and recognize distorted objects in image scenes. We describe a new way to design distortion-invariant correlation filters that ensures good generalization (same performance on training and test sets). The traditional way of designing correlation filters uses different types of frequency domain preprocessing and linear combination of training images. We show that these different approaches can be implemented in a framework using linear combination of eigen-images of preprocessed training data. Using eigen-domain data is shown to generalize well regardless of preprocessing used. We show results on SAR data using eigen-MINACE filters.

A problem of central importance in pattern recognition is the generation of an invariant response to a set of in-class objects while simultaneously maintaining discrimination against set of out-of-class objects, often together with resistance to background clutter. We apply a non-linear pre- processing operation to a wavelet filtered input and reference image prior to a correlation operation, i.e. in the space domain rather than to the spectrum. The performance of the filter is investigated by simulation for several different bandpasses and different degrees of non-linearity. The modified wavelet filter shows superior performance to the linear filter in terms of tolerance to in-class variations, discrimination ability and, most particularly, in its robustness to clutter in the input scene. This new operation was also employed in the synthesis of a modified synthetic discriminant function (SDF) filter, by applying the non- linearity to each of the wavelet filtered training set images comprising the SDF. The filter shows good detectability of an object in clutter, excellent discrimination ability without the need to include the out of class objects in the SDF, and good invariance to out of plane rotation over a distortion range of up to 90 degrees. The processing sequence is suitable for implementation by a hybrid digital/optical arrangement in which the input image is wavelet filtered in real-time by a DSP and a bipolar amplitude spatial light modulator employed to introduce the pre-filtered image into an optical correlator.

We present the derivation of an optimal correlation filter for fingerprint verification. The filter comprises multiple versions of the system user's fingerprint (i.e. it is a composite filter). Also, the characteristics of the filter can be adjusted so that its performance in a correlator is similar to that of a matched-filter or an inverse-filter, or some compromise between the two. It is these attributes that make this filter structure attractive for the task of fingerprint verification. The composite nature of the filter offers distortion tolerance by encompassing several different versions of the fingerprint image, while the tailored characteristics of the filters allows us to produce output correlation planes that can easily be processed. The filter was developed using a 'standard' database, with the objective of separating the two classes of input to the system: 'legitimate users' and 'attackers.' Specifically, the filter is optimized to minimize the probability of error (i.e. misclassification of user). Both the design and the implementation of the optimal fingerprint filter are covered in this paper.

Over the years many correlation filter designs for automatic target recognition have been proposed. Some designs offer high tolerance to different types of noise and clutter with the disadvantage of not so sharp correlation peaks. Other designs give sharp correlation peaks but low tolerance to disturbances. A fusion of different types of filters where the weaknesses of each design can be avoided and the strengths preserved has earlier been proposed. We have used an optical correlator with binary SLM:s. Because of this, the complex gray-scale correlation filters are binarized. The input image is edge enhanced and binarized before entering the system. By using different binarizations, the filters produce different false alarm-peaks, but equivalent correlation peaks in the correlation plane. Optical correlation is performed with the filters and the results are fused giving the resulting correlation image. Fusion of the results leads to suppression of false alarms, and enhancement of the true correlation peaks. Tests have been performed on both high and low contrast cluttered images with good results.

As we published in the last few years, when the given input- output training vector pairs satisfy a PLI (positive-linear- independency) condition, the training and the application of a hard-limited neural network can be achieved non-iteratively with very short training time and very robust recognition when it is applied to recognize any untrained patterns. The key feature in this novel pattern recognition system is the use of slack constants in solving the connection matrix when the PLI condition is satisfied. Generally there are infinitely many ways of selecting the slack constants for meeting the training-recognition goal, but there is only one way to select them if an optimal robustness is sought in the recognition of the untrained patterns. This particular way of selecting the slack constants carries some special physical properties of the system -- the automatic feature extraction in the learning mode and the automatic feature competition in the recognition mode. Physical significance as well as mathematical analysis of these novel properties are to be explained in detail in this article. Real-time experiments are to be presented in an unedited movie. It is seen that in the system, the training of 4 hand-written characters is close to real time (less than 0.1 sec.) and the recognition of the untrained hand-written characters is greater than 90% accurate.

A new nonlinear morphological detection algorithm is proposed for input scenes with a number of objects present in a clutter. It uses a synthetic discriminant function (SDF) to form the matched spatial filter (MSF) of the structuring element (SE) used in the hit-miss transform (HMT) detection. The SDF synthetic technique is used to adapt intraclass distortions and interclass similarity, where an HMT is used to adapt noisy and cluttered scenes. Simulation results show the proposed algorithm is an attractive nonlinear optical image processing technique that can be applied to an HMT detection to improve both the false alarm rate and its ability to detect multiple-object with distortions and clutter.

Automated processing and classification of real-time x-ray images of randomly oriented touching pistachio nuts is discussed. The ultimate objective is the development of a system for automated non-invasive detection of defective product items on a conveyor belt. This approach involves two main steps: preprocessing and classification. Preprocessing locates individual items and segments ones that touch using a modified watershed algorithm. The second stage involves extraction of features that allow discrimination between damaged and clean items (pistachio nuts). This feature extraction and classification stage is the new aspect of this paper. We use a new nonlinear feature extraction scheme called the maximum representation and discriminating feature (MRDF) extraction method to compute nonlinear features that are used as inputs to a classifier. The MRDF is shown to provide better classification and a better ROC (receiver operating characteristic) curve than other methods.

In many image processing applications, the field of view is not large enough to display the required scene. We present a method for reconstructing an entire scene from multiple, partially overlapping fields of view where the lateral position, orientation and the amount of overlap of the camera views are unknown. Rotation-invariant, correlation based filters are used to determine if two image segments (fields of view) register. The filter output provides the relative position and rotation between matching segments. All filters are created on-line. An automated supervisor selects image segments for correlation and then uses these results to register the image segments and assemble the entire scene without human intervention. Also presented are the results of software controlled automatic assembly of a multiple image scene.

A novel feature enhancement and similarity suppression algorithm is presented. This algorithm can be used for training matched filter based optical pattern recognition systems or for the correct recovery of noisy images from optical memories such as optical disks. The underlying theory and the results from computer simulations using human faces are presented. Some preliminary results have shown high probability of recognition in pattern recognition tests using the training data.

An efficient technique for restoring the original from a blurred noise corrupted image using the concept of fringe- adjusted joint transform correlation is proposed. This technique employs Fourier plane apodization using the reference image power spectrum and noise power spectrum to suppress the effects of blur as well as input scene noise. The performance of the proposed technique has been enhanced significantly by employing the Fourier plane image subtraction especially for blurred noisy input scenes involving multiple identical objects. The image subtraction technique eliminates the unwanted zero-order term and crosscorrelation terms produced by similar input scene objects while alleviating the detrimental effects of blur and noise that may be present in the unknown input scene. Computer simulation results using blurred noise corrupted grey level input scenes are presented to verify the performance of the proposed technique.

Application of visual image defocusing is illustrated as a technique for generalization, recognition and control. Specifically, formation of generalized images of objects by using a set of elementary patterns is described. This 'geometric alphabet' is created by two basic figures: a stripe and round spot. It is known from physiology that these figures produce response in special cells of the visual cortex of living organisms. Boolean algebra (Venn's diagrams) is used to obtain 'letters' and generalized images ('geometrical words') by computer simulation. The results of physical and computer experiments are given. A diagram of an anthropomorphic robot is presented together with model experiments on 'drawing' generalized images.

We present in this paper a study of the influence of the target resolution and contrast on correlation filter performance. Classical filters and the Optimum filter designed by Javidi et al. for non-overlapping noise were considered and tested on a set of realistic images based on real-world non- overlapping noise realizations. We show that for the images that we have tested, there is a limit of the resolution after which classical linear filters fail to detect properly targets, while the Optimum filter still performs well. Furthermore, it was observed that for some background images and fairly low target resolutions, there are values for the target illumination that prevent the classical filter to detect the target, while again it is possible to tune the Optimum filter to make it tolerant to varying target illumination. Eventually, we propose a new composite filter for in-plane rotation tolerance based on individual optimum filters and built according to the equal correlation principle perform best for our test images.

In this paper, a rotation invariant fingerprint identification system is implemented by using Dove prism. The input fingerprints are rotated with Dove prism and try out the correlation. We present that this system has the rotation invariant properties and can recognize the fingerprint in real-time. Binary phase only filter (BPOF) used for the spatial matched filter. Through some experiments, we also show that this system has a good performance in the rotated fingerprints.

In this paper the projection-slice SDF (PSDF) filter for rotation invariant response is discussed for pattern recognition applications. This method uses one half of a slice of the Fourier transform of the object to generate the transfer function of the filter. This is accomplished by repeating the half slice in the Fourier domain through 2(pi) radians about the zero-frequency point of the Fourier plane. This filter has the advantage of always matching at least one half of a slice of the Fourier transform of any rotation of the image. An analytical discussion of the filter transfer function and impulse response are presented along with simulated correlation results for a particular target scene including decoys. These results are evaluated using the peak correlation value, peak correlation energy, and the Fisher metric for performance.

A real-time, VanderLugt-type optical correlator with a single SLM has been developed. A field programmable gate array was used to capture and process images obtained from a CCD camera at a rate of 60 video fields per second. During both enrollment and verification, a finger slides over a glass prism and is input to the system via the frustration of total internal reflection. An auto-enrollment procedure captures the optimal image during each slide. An optimal composite filter is implemented. The correlation detection process comprises real-time tracking of the correlation peak while the finger is sliding, and a decision process based on projective decision boundaries. Real-life tests yielded a false rejection rate of 1% and a false acceptance rate of 0.2%.

Optimum filters for optical correlators have been the topic of several papers in the last few years. How can there be more than one optimum? Because, different approaches optimize different functions. Most research to date has optimized some internal working function of the correlator, such as optical efficiency, or narrowness of the correlation peak. However, in the real world, there is usually some application oriented function to optimize. For example, in a tracking system, the robustness to break-lock may be the most critical function. There is a threshold in the signal to noise required for single pixel tracking. Beyond that, 'optimization power' might better be spent on increasing tolerance to aspect change, rather than increasing signal to noise. This paper will discuss the attention the filter designer must pay to bore slope error, tolerances for distortions, signal to noise, correlation width, transport delay, and other variables.

In this paper we introduce the notion of a focused filter and discuss its application to the problem of pose estimation. A focused filter is a correlation filter designed to give a maximum response at one pose of the target. This pose is called the focus of the filter. As the actual pose of the target deviates from the focus, the filter's response should exhibit a graceful (and controlled) degradation. When presented with a test image, the responses of all focused filters are collected in a vector. This new vector will have a peak with the vector elements exhibiting the same shape as that used in designing one focused filter. This similarity is exploited for pose estimation by matching the filter responses to the designed shape. Simulation experiments are used to illustrate the potential of the new design method.

In this paper we report a study of optical pattern recognition using a cascade of linear shift invariant processing modules (correlators) each augmented with a thresholding layer. This configuration can be considered as a special class of multi- layer feed-forward neural network. In contrast with more generalized multi-layer networks, the approach is easily implemented in practice using optical techniques and consequently well suited to the analysis of large images. The concept of cascaded linear shift invariant processing is introduced within the context of network analysis. It is shown that the system is equivalent to a multi-layer network which is constrained to have a shift invariant output. The system has been modelled using a modified back propagation algorithm with optimization using simulated annealing techniques. The performance of the system has been compared to that of single layer correlators using a range of synthetic filters taken from the published literature. In particular we show that the noise tolerance of the cascaded system is increased relative to that of the minimum variance synthetic discriminant function (MVSDF). In addition we show that discrimination is enhanced considerably with respect to minimum average correlation energy (MACE) filters for the case of similar input images.

An image encoding scheme using the phase-contrast technique and a random phase distribution is proposed to encrypt images in phase masks. The robustness of the encoding is assured by the non-linearities intrinsic to the phase-contrast technique and the bandwidth of the random phase distribution. The advantage of this method, compared to the previous methods proposed, is the direct encoding of the image without any iterative calculation to generate the phase mask. This approach permits practical applications since the final phase mask could be implemented using thermoplastic plates and spatial light modulators (Slims).

The rapid classification and discrimination of images using spatial distribution of spectral information has widespread applications from remote monitoring of vegetation and pollution damage, to military surveillance and anti-stealth warfare. This paper describes the construction of a practical optical correlator capable of spatial and spectral pattern recognition. The method used here exploits the ability of optical correlators to process information in parallel with high space bandwidth product. For reasons of light efficiency and practical convenience, the spectral information is input into the correlator as a spatial array of coherence or 'white light' fringe patterns. This technique we have called coherence transform imaging (CTI). This paper discusses the relative merits of several interferometric methods to perform CTI including Michelson, Sagnac and Polarizing interferometers. A robust CTI camera utilizing a polarizing interferometer is then described and simple matched filtering operations are performed using CTI images recorded on photographic film. Finally an optical correlator capable of real time spectral discrimination and tracking of colored objects is demonstrated.

Ultra fast imaging and destruction of the gall bladder stone is performed using Ho:YAG laser. A laser guided approach for lithotropsy is proposed. The correlation output peak is introduced as a feedback signal for firing the laser pulse for stone destruction and 'discrimination' of the tissue image so that the risk of damaging and perforation of the tissue is reduced. A system constituted by correlation of ballistic images and fluorescent signals is proposed.

As the necessity of automatized system has been increased, an adaptive target detection algorithm is widely demanded in the development of industrial and military technologies. Accordingly, this paper proposes a novel approach to detect and segment the moving targets from a sequence of images which are detected by a detector embedded on a moving vehicle such as car, airplane, ship, and so on. The BPEJTC (binary phase extraction joint transform correlator) that can produce phase type correlation peaks in real time by optical way, is introduced for compensating the camera movement between two successive frames.

This review paper concerns with the automatic recognition of Arabic characters by computers. A comprehensive survey of the field is presented with critical evaluation. Problem areas are identified together with proposed approaches for solutions. Finally, areas of further research are briefly discussed.

This paper will discuss the application of optical disk technology to optical processing. The limiting factor in optical image processing is the spatial light modulator. Spatial light modulators are currently limited by pixel size and TV frame rates. Optical disks provide pixel sizes on the order of a micron, fast access times and 650 MB storage space resulting in approximately 60,500 256 by 256 pixel images. CD, CD-Recordable, Magneto-Optic and Phase Change media have been studied. CD and CD-Recordable media provide high contrast, while phase-change media provides moderate contrast and rewritability. This paper will focus on an optical system that uses a phase-change disk as the filter medium in an optical correlator. Specially developed software to write two dimensional images to the phase-change disks will be addressed.

BMW Rolls-Royce is developing the civil turbofan engine family BR700 with a thrust range from 12.000 to 23.000 pounds. To improve the efficiency, life time and cycles of an engine it is necessary to know the wall temperature on its components. The distribution of turbine wall temperature is measured with so-called Thermal Paints, which change their color irreversibly according to the local maximum temperature. A CCD color line scan camera system to record and analyze these data is being developed. To understand the thermal distribution across the surface of a component, the paint color change boundary is identified and transformed into calibrated temperature data. This thermal distribution can be used to validate the engineering thermal model.

A new type of information processing system is developed. The processing is implemented by using hybrid architecture of Vander-Lugt optical correlator and joint transform correlator. In one optical setup, joint transform correlator is used for object tracking and Vander-Lugt optical correlator is used for object recognition respectively. The experiments show that we can get good results.

In this paper, we develop an optoelectronic pattern recognition system based on an incoherent optical correlator. The real-time input image is correlated with stored database images consecutively after several necessary preprocessing steps, and the recognition result is given quickly. The key device for the optical correlation is a SHARP QA-1200 8.4' active matrix TFT liquid crystal display (LCD) panel which is used as two real-time spatial light modulators (SLMs) for both the input and the database image. Several practical difficulties when utilizing the LCD panel as a light intensity modulator have been overcome. Experimental results are given.

The problems of investigation and calculations results reliability (CRR) evaluation in holographic computing systems are considered. The analysis of the existing approaches to estimation of CRR has been overtaken. A new method of CRR is proposed. The evaluation of CRR in the systems with single and coded correlation responses regarding processing of the images and the images Fourier spectrum has been performed. The simulation results are submitted.

We showed that advanced distortion-invariant filters such as OTSDF and DCCF filters can be used in conjunction with circular filters to obtain rotation-invariant correlation filters. The combination allows the calculation of OTSDF and DCCF filters to be greatly simplified when compared to using rotated views of an object to create a filter. The performance of the filters were different yet overall comparable indicating that perhaps different training data could have altered the results. In addition, the filters generated were real-valued so they may be implemented on a wider variety of SLMs usually associated with general distortion-invariant filters.

The Optical Processor Enhanced Ladar (OPEL) program is designed to evaluate the capabilities of a seeker obtained by integrating two state-of-the-art technologies, laser radar, or ladar, and optical correlation. The program is a thirty-two month effort to build, optimize, and test a breadboard seeker system (the OPEL System) that incorporates these two promising technologies. Laser radars produce both range and intensity image information. Use of this information in an optical correlator is described. A correlator with binary phase input and ternary amplitude and phase filter capability is assumed. Laser radar imagery was collected on five targets over 360 degrees of azimuth from 3 elevation angles. This imagery was then processed to provide training sets in preparation for filter construction. This paper reviews the ladar and optical correlator technologies used, outlines the OPEL program, and describes the OPEL system.