We consider the use of new distortion-invariant optical correlation filters for machine-printed OCR. Our work is unique in its treatment of a large set of different fonts, printer types, plus rotations and scale (point size) variations, and various practical issues such as printing artifacts and background noise. We detail their use in the locations and recognition of alphanumeric fluids (digits) in destination address blocks (DABs).

The conventional Synthetic Discriminant Function (SDF) filters are complex-valued and thus cannot be accommodated on spatial light modulators that can represent only a subset of all possible values in the complex plane. Here, we compare the performance of different SDF filters designed to satisfy certain device restrictions.

An innovative optically implemented morphological processor is introduced. With the use of a large space-bandwidth-product Dammann grating and a high-speed shutter spatial light modulator, effective structuring element with large size and arbitrary shape can be constructed with dynamic reconfigurability. This reconfigurability is a major improvement over the conventional correlator-based morphological processor in which fixed holographic filters are used as structuring elements (Casasent and Botha, 1988). A novel two-dimensional thresholding photodetector array, capable of performing parallel thresholding and feedback, is utilized in this system and makes possible the implementation of many complex morphological operations requiring iterative feedbacks and full programmability. The optical architecture and the principle of operation are presented. Experimental demonstration of binary image morphological erosion, dilation, opening, and closing are also demonstrated. A technique for extending this technique to gray-scale image using thresholding decomposition technique is also discussed.

The design of real-valued composite filters for optical pattern recognition and classification is considered. A procedure to design a real-valued minimum average correlation energy (MACE) filter is developed. Also, the design of a real MVSDF-MACE filter that minimizes the output variance due to input noise while maintaining a sharp correlation peak is developed. Computer simulation indicates that the performance of these real filters is almost as good as that of the complex filters.

The correlation performance of nonlinear joint transform correlator for input signals containing a target in the presence of the input scene noise that is disjoint with the signal is investigated. It is shown that the binary joint transform correlator outperforms other types of nonlinear joint transform correlator for lowpass scene noise. Correlation tests using computer simulations are presented. The nonlinear joint transform correlator performance is determined for various degrees of nonlinear transformations and different input scene noise parameters. The investigations indicate that the correlation performance is sensitive to the bandwidth variation of the small bandwidth noise and is less sensitive to that of the wide band noise.

We report that nonlinear parameters effect the characteristics of the optical pattern recognition of a feedback joint transform correlator (FJTC) which is based on a joint transform correlator (JTC). This system incorporates the optically addressed ferroelectric liquid crystal spatial light modulator named LAPS-SLM. Nonlinear parameters in the FJTC are a threshold level of the LAPS-SLM to record the binary joint power spectrum (JPS), and a feedback transfer function from the correlation plane to the input plane. It is important to control the threshold level of the binary JPS. But the FJTC does not always need the optimum condition where the FJTC recognizes the signal image correctly at the initial recognition.

Adaptive Resonance Theory provides a neural network architecture for self-organizing arbitrary input patterns into stable categories. In our work we utilize a model of ART known as ART2-A, which is capable of processing both analog and binary patterns. Our model is adapted to handle 2-D images for input patterns, and to allow for translation invariance. The computation of image patterns is assisted by a joint transform correlator (JTC), providing a fast, real-time, translation-invariant method of initial comparison. The JTC has an advantage over other optical correlator architectures in that a separate matched filter for each input need not be constructed. In this paper, we present a brief overview of the ART2-A algorithm and our optoelectronic implementation of this neural network model. This paper is based on work by Kane and Paquin submitted to IEEE Transactions on Neural Networks entitled 'POPART: Partial Optical ImPlementation of Adaptive Resonance Theory 2'.

We present a new optoelectronic quasi phase processor based on a conventional joint transform correlator architecture and using a liquid crystal SLM. The phase information is extracted from a joint transform spectrum producing a good correlation performance in terms of noise tolerances and peak sharpness. Several characteristics of the used SLM improving the correlation function are investigated. Computer simulations of the correlator performances and experimental results of an application on tool recognition are presented and discussed.

It has been shown that tracking small particle motion can be accomplished by tracking the speckle pattern is produces. This paper describes various methods of real-time tracking of speckle patterns obtained from ultrasonic flow imaging of blood and tissue motion using optical correlation. Results obtained from a gray scale joint transform correlator utilizing a twisted nematic liquid crystal spatial light modulator and from Sandia Lab's acousto-optical correlator are presented. The experimental results demonstrate the feasibility of real-time tracking with accuracy comparable to that of template matching algorithms currently being implemented in digital hardware.

The joint transform correlation can be optimized by subtracting the constant terms (Fourier transform power spectra) in the filter plane. Computer simulations for three approaches to optimization have been performed and have shown that the contrast of joint transform power spectrum can be maximized to produce optimal correlation signal.

Some of the problems in realizing a practical optical correlator based ATR system are discussed. Two factors play a key role in the discussion: (1) the system and component issues of optical correlators, and (2) the processing potential of digital processors. The lack of an appropriate Filter SLM is identified as a major problem area for optical correlators, both now and in the future at the current rate of development and emphasis.

Optimal trade-off filters for the Signal-to-Noise Ratio, the sharpness of the correlation peak and the optical efficiency are discussed. A unified formalism is introduced for Circular Harmonic (CH) and Synthetic Discriminant Function (SDF) filters. With this formalism, new solutions for CH and SDF filters are derived.

Replication in the output plane of an optical correlator, due to pixelation of the Fourier plane filter, can lead to false correlation signals. This paper suggests randomization of the pixel positions of the Fourier plane filter as a solution, and demonstrates its effectiveness through computational simulations and optical correlation experiments using a custom made SLM.

We describe a complex optical system consisting of a 4f optical correlator with programmable filters under control of a digital on-board computer that operates at video rates for filter generation, storage, and management. It gives intelligent vision to a semi-autonomous vehicle, with ability to recognize immediate danger to its survival in the near term and ability to pursue navigational goals on the basis of tracking the previously identified features.

In a conventional pattern recognition system with a matched spatial filter (MSF), the system is based on a template matching which is the comparison of the test pattern with a number of stored patterns until an exact match is found. Therefore, sometimes, the system has not been able to distinguish some similar patterns. In this paper, we propose to apply a concept of feature extraction to matched spatial filtering to improve performance of the MSF for distinction of some closely similar patterns. The synthesis of the MSF with partial patterns of unknown input objects to be recognized enables filtering operation to extract desired features from some input patterns. Using this technique and a hybrid pattern recognition system with the MSF synthesized by nine feature extracted patterns, we could perfectly recognize a page of unknown 25 alphabets. The recognition result was given not by conventional peaks but by character symbols on a CRT display.

We provide insight into the disadvantages of various non-linear distortion-invariant optical correlation filters. From this, guidelines for improved optical correlation filters emerge including: filters for intra-class recognition, filters for clutter rejection and the different types of clutter that arise, hierarchical inference filters, remarks of Fourier vs. image domain synthesis, filter space-bandwidth product (SBWP), techniques to reduce correlation plane energy and filter performance measures and issues.

Use of the Magneto-Optic Spatial Light Modulator (MOSLM) in the filter plane of optical correlators is attractive because it is capable of high frame rates. This has led to the design of several different types of Binary Phase-Only Filters (BPOFs) and Ternary Filters matched to a single reference. To further increase the throughput of an optical correlator employing the MOSLM, attempts at designing Binary Phase-Only Composite Filters have been made. In this paper, we introduce a new iterative technique to design such Composite Filters in which we simultaneously maximize the peak sharpness and minimize the deviation of the actual correlation peak intensities from their desired values.

Experiments on k^th law nonlinearly transformed matched filters for optical correlation are provided. Experimental results for the images tested indicate that nonlinear matched filters produce good correlation performance in the terms of correlation peak intensity, signal-to- noise ratio, and peak-to-sidelobe ratio. The sensitivity of the k^th law nonlinearly transformed filter to rotational changes of the input signal is investigated. It is shown that for the images presented here, up to a certain degree of input signal rotation, a highly nonlinearly transformed matched filter may produce a larger peak-to-sidelobe ratio than a conventional matched filter.

The ternary phase-amplitude filter (TPAF) is by definition restricted to the modulation values -1, 0, and 1, thus comprising a binary phase-only filter (BPOF) multiplied by a binary- amplitude pattern, i.e., a region of support. The TPAF offers an attractive combination of real-time implementation with available devices and good correlation performance. Smart (optimized distortion-invariant) TPAF formulations have been developed. The TPAF enables filter implementation with magneto-optic devices and these devices also can be used for image input if gray scale scenes can be binarized while preserving good correlation performance. We provide simulation results addressing the comparative performance of mixed-metric smart TPAF's using gray scale, edge-enhanced and binary images derived from identical original scenes. The variation of filter performance with training set background intensity level is examined.

In an optical correlator, binary phase-only filters (BPOFs) that recognize objects that vary in a nonrepeatable way are essential for recognizing objects from actual sensors. An approach is required that is as descriptive as a BPOF yet robust to object and background variations of an unknown or nonrepeatable type. We developed a BPOF that was more robust to unknown variations than a binary version of a synthetic discriminant function (fSDF) filter. We compared the values of spatial frequencies of a training set and compared them in terms of their similarity. Then, we grouped them into a cluster by forcing some values to zero. In this way, we retained the invariant spatial frequencies of a training set and generated a ternary filter. Our filter offered a range of performance by adjusting a parameter. At one extreme, our filter offered similar performance to that of a fSDF filter. As the value of the parameter was changed, correlation peaks within the training set became more consistent and broader as the filter became more robust. In addition, the feature-based filter was potentially useful for recognizing objects outside the training set. Furthermore, the feature-based filter was more easily calculated and trained than an fSDF filter.

We consider wavelet and Gabor transforms for detection of candidate regions of interest in a 2-D scene. We generate wavelet and Gabor coefficients for each spatial region of a scene using new linear combination optical filters to reduce the output dimensionality and to simplify post-processing. We use two sets of wavelet coefficients as indicators of edge activity to suppress background clutter. The Gabor coefficients are found to be excellent for object detection and robust to object distortions and contrast differences. We provide insight into the selection of the Gabor parameters.

The advantages of a wavelet transform over a short time Fourier transform are explained using 1-D synthetic data in a computer simulation. An optical set up is described in which an electronic analog sensor signal drives an acoustooptic cell that controls the intensity of an Argon laser. A mechanical scanner writes the information as a line onto a spatial light rebroadcaster (SLR) module containing an optical liquid crystal light valve. A lens system expands the line into a 2-D array. A wavelet transform filter is placed in the Fourier transform plane of a 4f correlator. An optical demonstration shows the formation of a wavelet transform and an inverse wavelet transform to reconstruct the original waveform.

Ring-wedge detectors are known to produce a useful feature set for certain types of pattern recognition. Their major shortcoming is that they measure global features. We present an optical processor, based upon the computation of a two-dimensional wavelet transform, which overcomes this limitation. By using wavelet functions that are essentially compact in the space domain we generate an output that consists of a mosaic of spatially localized bandpass components. Consistent with the nature of wavelets, the radial (ring) frequency components, are organized into constant-Q (f₀/(Delta) f) bands. The angular (wedge) frequency content is divided into a number of equal-width bands which cover the full 0 - 180 degree(s) range. The radial information is obtained by using a feedback iteration loop which scales the input image by a fixed factor for each time around the loop. By introducing a tilt in the Fourier plane of the feedback loop, we arrange for the space-domain representations of each scaled input to be spatially separated without altering the position of their Fourier transforms. In this way, all of the radial frequency bands can be extracted with a single wavelet filter. A diffraction grating is introduced into the optical path after the scaling loop to replicate the Fourier information M times. The angular information is extracted using M filters in parallel-- one for each wedge component. The filter outputs are minified consistent with Nyquist theory for their reduced bandwidths to produce an output whose space-bandwidth product is roughly the same as that of the input image. We present the design of the optical system along with some initial experimental results.

Digital and optical implementations of the holographic ring detector directly coupled into an optical neural network are studied as a means of increasing the speed of the decision process for particle characterization. A single holographic optical element which performs the same function as a ring detector and a simple two-layer, feedforward optical neural network is fabricated and evaluated.

One-dimensional hologram technology for parallel information recording on a continuously moving carrier in both thin and thick photolayers is considered. The possibility of superposing 1-D holograms allowing the information capacity to be increased up to several Gbytes on one side of the disk is shown. A high-speed holographic disk memory is disclosed. The application for designing neural nets with serial- parallel processing is discussed. For a two-layer neural net, the effective number of neurons is of order 10⁵ with 16 real neurons in each layer.

This paper describes two processing algorithms that can be implemented optically: the Radon transform and angular correlation. These two algorithms can be combined in one optical processor to extract all the basic geometric and amplitude features from objects embedded in video imagery. We show that the internal amplitude structure of objects is recovered by the Radon transform, which is a well-known result, but, in addition, we show simulation results that calculate angular correlation, a simple but unique algorithm, which extracts object length, width, area, aspect ratio, orientation and boundary from suitably thresholded images. In addition to being insensitive to scale and rotation, these simulations indicate that the features derived from angular correlation algorithm are relatively insensitive to tracking shifts and image noise. Some optical architecture concepts, including one based on micro-optical lenslet arrays, have been developed to implement these algorithms. We will discussed these architectures, stressing the micro-optical approach. Test and evaluation using simple synthetic object data will be described. We will also describe the results of a study that uses object boundary (derivable from angular correlation) to classify objects using a neural network.

We describe and present experimental results of the optical calculation of potential field maps suitable for mobile robot navigation. The optical computation employs two write modes of a microchannel spatial light modulator (MSLM). In one mode, written patterns expand spatially, and this characteristic is used to create an extended two dimensional function representing the influence of the goal in a robot's workspace. Distinct obstacle patterns are written in a second, non-expanding, mode. A model of the mechanisms determining MSLM write mode characteristics is developed and used to derive the optical calculation time for full potential field maps. Field calculations at a few hertz are possible with current technology, and calculation time vs. map size scales favorably in comparison to digital electronic computation.

Optical cellular logic and extended-interconnect cellular logic processor architectures are described. Simulations, tolerance studies and circuit construction are used to advance both device technology and architectural thinking. Benchmarking against electronic distributed array processors leads to an optical non-locally interconnected, off-chip scheme with local electronic connections between smart pixels, unless ultra-low switching energy all-optical components can be fabricated. Fundamental limits to optical bistability do not rule out the latter possibility.

There are at least two major classes of computers in nature and technology: connectionist and selectionist. A subset of connectionist systems (Turing Machines) dominates modern computing, although another subset (Neural Networks) is growing rapidly. Selectionist machines have unique capabilities which should allow them to do truly creative operations. It is possible to make a parallel optical selectionist system using methods describes in this paper.

A novel microchannel plate (MCP) based electrooptical A/D converter is proposed for signal bandwidths in excess of 100 MHz. The signal forward-biases an electron flux through 500 impedance matched transmission lines deposited on an MCP. The backbiasing of the gap between adjacent MCPs provides a voltage threshold for the signal triggering of an electron cascade in the second MCP. The second MCP is biased by a square wave timing pulse in order to measure the time at which the signal exceeded the backbias threshold. It is shown that reasonable MCP operating parameters result in 100 psec time resolution. The Nyquist criteria is applied to determine the least bit size for voltage sampling of a transient pulse.

We describe an optical, scanning-based imaging system for directly acquiring the directional gradient of an image. We later discuss a modification to the system which enables it to track contours of moving objects. We present results of proof-to-principle experiments and discuss applications of the system to the identification of objects.

Piston-only spatial light modulators, such as the flexure-beam deformable mirror device, hold great promise for real-time optical processors because of their ability to accurately match the phase of arbitrary laser images. Non-ideal devices may have uncontrolled phase errors which can limit performance. A statistical analysis has been developed which models the effect of random piston and tilt errors on the diffraction pattern of phase-only SLMs. A slight modification of these equations describes the performance of these SLMs in the phase-only correlator and the phase-only correlator modulated by pseudo-random sequences. Results on correlator diffraction efficiency versus amount of phase error are presented. For the specific case of a binary phase-only matched filter a diffraction efficiency of 40.5% is found.

An efficient method based on the symbolic substitution technique is proposed for the conversion of modified signed-digit numbers to binary numbers. As a specific example, the conversion of the output of an 8-bit adder with the proposed method is investigated and compared with a previous method.

The Index Interferometer is a novel instrument being developed by Northeast Photosciences. The instrument is a breakthrough in the high-accuracy measurement of the index of refraction, the dispersion, and the index profile of materials. The instrument accurately measures the index of refraction of materials to one or two more significant figures than previous instruments. Material slices polished moderately flat are sufficient, without any requirement for special or complicated material shapes, such as prisms. The index profile at any chosen wavelength can be measured using a simple color filter. No special laser sources or carefully collimated parallel beams are required. The index profile over an entire sample can be directly obtained at any desired wavelength. This instrument is remarkable in that it greatly increases the accuracy of measurement, eliminates the need for high-quality, extremely narrow sources and for fabrication of special-geometry samples, and adds additional features, such as index profile measurements. The technique compares the fringe pattern from the top surface with that from a reference mirror to determine the thickness. Then, with the aid of a filtered white light source, the interference pattern from the back surface is compared with that from the front to yield the optical thickness of the sample. The combination of the two measurements gives the index. The back surface fringe pattern itself gives the index profile.

Speckle interferometry provides a basis for analyzing the dynamic motions of materials through differencing of the field scattered from objects at two different times. Phase differences in the signals measured at different times, inferred from fringe patterns, indicate the degree of deformation present. Automatic analysis of differences images requires significant preprocessing to enhance the contrast of fringe regions. Often fringes that are evident to the human eye cannot be perceived automatically because the fringes usually consist of widely separated high intensity spikes. Median or averaging filters are ineffective at enhancing these patterns. Adaptive filtering similar to that used in SAR image analysis is capable of enhancing the fringe area contrast. A two step process is detailed. In a first phase a filter based on window mean and variance suppresses noise and generates a greater cohesion of high intensity points in the fringe areas. In a second phase the image is average filtered to smooth the intensities in the fringes. An auxiliary routine used to count fringes is discussed. Comparisons with median filtered results show the greater ability to automatically count fringes using this two step method.

A unified approach for a general metric that encompasses both the signal-to-noise ratio (SNR) and the peak-to-correlation (PCE) ratio in optical correlators is described. In this approach, the connection between optimizing SNR and optimizing PCE is achieved by considering a metric in which the central correlation irradiance is divided by the total energy of the correlation plane. The peak-to-total energy (PTE) is shown to be optimized similarly to SNR and PCE. Since PTE is a function of the search values G and beta, the optimal filter is determined with only a two-dimensional search.

A real-time 2-D angular-multiplex beam-array holographic storage and reconstruction technique using electrically-addressed spatial light modulators(E-SLM's) and photorefractive crystals is described. Using a liquid crystal television (LCTV) spatial light modulator (SLM) for beam steering and lithium niobate photorefractive crystal for holographic recording, experimental results of generating large and complicated arrays of laser beams with high diffraction efficiency and good uniformity are presented.

Many applications of imaging systems require information about the spectrum of light that forms the scene. One way of deriving spectral information from a source is by calculating the Fourier transform of the temporal coherence function, as measured by a Michelson interferometer. This is the basis of Fourier transform infrared spectroscopy (FTIR). With a phase-shifting spatial light modulator, it should be possible to construct a two-dimensional array of Michelson interferometers in order to derive independent spectral estimates for each pixel in a natural light image. Realistic constraints on the range of motion of mirror pixel elements will limit the spectral resolution of such a system, but these limits also allow the use of wide aperture imaging optics, and assure that the data processing burden is within the range of current digital technology.