In general, a decision on any event is made through a sequence of bifurcating selection process. In addition to numerical computation, the binary logic operations of digital electronics may well be used to describe any complicated decision making procedure. In this paper, we show that the single-input-double-output bifurcating principle may be applied to optical information processing. In particular, the massive parallelism and inherent inaccuracy of optics offer a unique representation of human thinking and decision making process. Coherent optical experiments including pattern recognition and dynamic range compression via photorefractive crystals are used to demonstrate the principle of bifurcating optical information processing.

One of the important aspects of a neural network is the exploration of the spatial content of the object under observation in this connection. We shall present a color optical neural network using inexpensive package-size liquid crystal televisions (LCTVs). By introducing a color encoding technique in conjunction with a conventional white light source, a multicolor neural net is synthesized. We have shown that by exploiting the spectral component of the LCTVs storage capacity of the neural net can be improved. Simulations as well as experimental results obtained from the proposed LCTV color neural net are provided, in which the effects due to noise and due to color crosstalk are addressed.

Neural nets present various possibilities for local and global processing in digital halftoning. For global processing the Hopfield-type network was examined for its usage. The constraints concerning this model are presented and their effects for the binarization problem are shown. The numerical description leads to a basic parallelism to the well known iterative Fourier transform algorithm (IFTA), applied in digital halftoning.

Training by adaptive gain (TAG) neural network model, which had been developed for optical implementation of large-scale artificial neural networks, is further extended for better performance and its feasibility is demonstrated by a small-scale electro-optic implementation. For fully interconnected single-layer neural networks with N input and M output neurons the modified TAG model contains two different types of interconnections, i.e., MN fixed global interconnections and (beta) N + M adaptive local interconnections. For the original TAG model the number of adaptive local interconnections (beta) was set to 1, and the interconnections were understood as adaptive gain. For 2-dimensional input and output patterns the fixed global interconnections may be achieved by page-oriented holograms, and the adaptive local interconnections by spatial light modulators. The original and modified TAG models require much less adaptive elements than the popular perceptron model with fully adaptive global interconnections, and provide possibilities of implementing large-scale artificial neural networks with some sacrifice in performance. The training algorithm is based on gradient descent and error back-propagation, and is easily extensible to multi-layer architecture. Computer simulation and electro-optic implementation demonstrate much better performance of the modified TAG model compared to the original TAG model.

We have proposed and demonstrated a shift-invariant optical associative memory by the modified Hausler/Lange algorithm. It is a feedback network with space invariant coupling by the convolution operation. The system is shown to have the capabilities of error correction and pattern recognition. Both computer simulations and optical experiments are presented in this paper.

We describe a two-layer neural network using holographic optical disks as the interconnection weights. Such a system can be used to implement one two-layer with a large number of hidden units, or several two-layer networks with a smaller number of hidden units

We present a color exemplar-based neural network that can be used as an optimum image classifier or an associative memory. Color decomposition and composition technique is used for constructing the polychromatic interconnection weight matrix (IWM). The Hamming net algorithm is modified to relax the dynamic range requirement of the spatial light modulator and to reduce the number of iteration cycles in the winner-take-all layer. Computer simulation results demonstrated the feasibility of this approach

Using an electron-beam addressed spatial light modulator (EBSLM), an optical three-layer neural network with the back-propagation learning algorithm is successfully implemented. The weight matrices are memorized and modified on the EBSLM. To realize a large number of learning iterations, we propose a new renewal method of the weights.

A perfectly convergent, unipolar, neural associative memory system based on nonlinear, dynamical terminal attractors is presented. By adaptively setting the threshold values for the dynamic iteration for the unipolar, binary neuron states with terminal-attractors, the achievement of perfect convergence and correct retrieval has been demonstrated via computer simulation. The simulations are completed by (1) exhaustive tests with all of the possible combinations of stored and test vectors in small-scale networks, and (2) Monte Carlo simulations with randomly generated stored and test vectors in large-scale networks with an M/N ratio equals 4 (M: the number of stored vectors; and N: the number of neurons up to 256). An experiment with exclusive-or logic operation using LCTV SLMs is used to show the feasibility of an optoelectronic implementation of the model. The behavior of terminal attractors in basins of energy space is illustrated through examples.

A learning-based image recognition system using a thin photorefractive crystal plate is presented and demonstrated. The matched filter is synthesized by the modified perceptron learning network and is recorded in a 100 micrometers thick LiNbO₃ crystal plate. The system performs a shift-invariant correlation for image recognition.

After a rapid introduction that presents the main reasons for the renewal of interest for photorefractive holographic memories, we first discuss their potentialities and limitations. We then continue by a presentation and analysis of novel techniques that permit considerable facilitation of the memory operation and improve its capabilities. In conclusion, we indicate envisioned applications for photorefractive holographic memories.

We consider some fundamental limits to the volume holographic storage in photorefractive media, including the storage capacity and the dynamic range. We also discuss the photorefractive noise gratings formed during a multiple exposure schedule. These fundamental issues in photorefractive storage are important in understanding the potentials and limitations of optical memory systems.

We describe a photorefractive 3-D optical disk for data storage. The information is stored at several recording locations on the disk as angle-multiplexed holograms. Design parameters that affect alignment sensitivity and diffraction efficiency are considered.

A photorefractive strong volume hologram by using anisotropic diffraction in a normal-cut BaTiO₃ is proposed and demonstrated. The potential applications of this strong volume hologram are discussed.

A joint transform correlator (JTC) is presented in this paper that utilizes a thin LiNbO₃ crystal plate in the Fourier plane. The crystal plate shows the real-time dynamic photorefractive characteristics. We first report investigations on the planar grating properties of a thin Fe-LiNbO₃ crystal plate and then demonstrate a new JTC system by this plate. The system has the capability of shift-invariance and real-time adaptive filtering.

This paper reviews the recent progress in the development of holographic neural networks using surface-emitting laser diode arrays (SELDAs). Since the previous work on ultrafast holographic memory readout system and a robust incoherent correlator, progress has been made in several areas: the use of an array of monolithic `neurons' to reconstruct holographic memories; two-dimensional (2-D) wavelength-division multiplexing (WDM) for image transmission through a single-mode fiber; and finally, an associative memory using time- division multiplexing (TDM). Experimental demonstrations on these are presented.

In this paper, we present a new opto-neural approach to the problem of multi-target tracking. The proposed hybrid opto-neural system uses an optical joint transform correlator to reduce the massive input target data into a few correlation peak signals and then a massive parallel computational neural network algorithm is used for effective target tracking data association based on these correlation signals. For real-time operation, a nonlinear joint transform correlator is optically implemented using a high resolution LCD spatial light modulator (SLM) and a new track based on field (TBF) neural network tracking algorithm is introduced to tackle the effective multi-target data association in a real-time basis. Through the computer simulation, the performance of the proposed hybrid opto-neural tracking system is evaluated and some experimental results on simultaneous tracking of multi-targets are also provided.

A fiber speckle-hologram is formed by two coherent light fields, in which a speckle field emerging from a multimode fiber combines with a convergent reference beam. When the speckle-hologram is illuminated by the same speckle field from the fiber, a high intensity correlation peak is reconstructed. If the fiber is subjected to some status changes, the correlation peak gradually decreases and disappears. The main objective of this paper is to use a `core-ring-intensity-ratio' technique to analyze the correlation peak so the fiber status changes can be determined. An experimental demonstration to confirm this relationship is also given.

In this paper we propose a feature enhanced interpattern associative (FEIPA) optical neural network. The common part of the stored patterns is regarded as redundance and its contribution in the association process is discarded. Therefore, the output before thresholding is more uniform, and hence, it is easier for the thresholding performance and increases the iteration speed. Furthermore, the optical implementation is much easier because all the elements of the interconnection matrix are non-negative and unipolar. The theoretical description and the experimental results are presented.

This paper reviews some of the digital optical cellular image processing techniques that have been proposed or experimentally demonstrated.

A novel optical free-space mesh-connected bus (MCB) interconnect network architecture is proposed. An MCB is known to have the capability of interconnecting, using a three-stage switching, N nodes with a power distribution loss proportional to (root)N, and is therefore advantageous for networking a large number, say over 1000, of communicating ports. Based on conventional space-invariant optical components in a compact and efficient geometry, the proposed optical MCB system concept can be used to build either free-space optical interconnect links for parallel processing applications or central switching systems for local or global lightwave communication networks. The proposed architecture lends itself for networking under both the wavelength-division multiple-access (WDMA) and other multiple- access environments. In this paper, based on the WDMA environment, various optical system implementation and performance issues are discussed and parameters are analyzed. It was found that using a reasonably compact three-dimensional free-space volume, more than 100,000 dispersion-limited communication nodes at a uniform channel spacing of 0.75 nm can be linked with a moderate PDL of 28 dB. Some preliminary optical WDMA MCB experiments based on a 27 X 27 panchromatic optical source array were performed to confirm the operational principle of the proposed concept.

Based on dual-rail logic, we propose a cascadable binary logic gate module in this paper. All 16 logic functions of two binary input patterns can be implemented. Experimental results are also presented.

We present a holographic array illuminator based on the Talbot effect. This illuminator can be used in massively parallel architecture with a good efficiency and minimal aberrations in the reconstruction at a different wavelength.

A simple way to synthesize a shift, fully rotational and limited size invariant composite matched spatial filter for a coherent optical correlator is proposed. We use circular harmonic components of the same order of a reference pattern in different sizes as the training images instead of using the whole reference pattern in different orientations and different sizes as is reported by other workers. This saves much labor and time in the synthesis of the filter. In this article, we have synthesized a simple filter containing four second order circular harmonic components of the alphabetic letter `E' in four relative sizes 1, 1.17, 1.33, and 1.5. Results of computer simulation have shown that this filter is shift, fully rotational and limited size invariant over the size range from 1 to 1.5. Computer simulation has also shown that this filter possesses discriminating ability.

In this work, a distortion-invariant pattern recognition scheme called the composite training image method is introduced. Usually, in attempting to detect the distorted (rotated, size- changed, shifted) versions of an object, a large number of raw training (distorted) images are used. However, there is a trade-off between this number and the ratio of signal correlation intensity peak to the maximum sidelobe (RSMS). In order not to degrade this ratio, the number of training images should be reduced as much as possible. We show how to fuse several similar raw training images into a composite training image. In this paper, we illustrate the feasibility of using such composite training images.

Theoretical expression and convergence conditions for the output of the feedback holographic associative memories are investigated in this paper. Feedback holographic associative memories are shown to reduce to pinhole sampling holographic associative memories under certain conditions. Therefore, the crosstalk in the output of feedback holographic associative memories can be minimized. With the theoretical results obtained in this paper, feedback holographic associative memories can be designed efficiently. In addition, feedback holographic associative memories are shown to be generalized holographic associative memories.

The principle of the optical joint transform correlator (JTC) is reviewed. The architecture using liquid crystal TV and CCD is discussed in detail. Various applications and optical architectures of JTC are summarized.

A four-wave mixing (FWM) optical correlator created by using a thin photorefractive (PR) bismuth silicon oxide (BSO) crystal plate is demonstrated. The envelope function which limits the shift-invariant range of the volume holographic correlator is derived. The optimum conditions for obtaining a compact shift-invariant FWM correlator are described. Both experimental and simulated results are presented.

Self-pumped phase conjugation is realized in Ce-doped (K_o.5Na_0.5)_0.2(Sr_0.61Ba_0.39) Nb₂ 0₆(KNSBN) crystal by using two beams of He-Ne laser, the self-pulsation and optical chaos due to the coupling of the beams are observed and studied.

A gray level discrete associative memory (GLDAM) neural network using interpattern association (IPA) model is presented. By decomposing a gray level pattern into bipolar/binary modes of subpatterns, a GLDAM can be constructed. Although GLDAM improves the information capacity of the neural net, the decomposition process introduces sparse allocation in memory matrix, which affects the performance of the neural net. Computer simulated results for the Hopfield and the IPA models are provided, in which we have shown that the IPA GLDAM performs better.

Experimental results of 2-D image addition and subtraction using self-imaging (Talbot) effect, without imaging elements and in real time, are presented. From them, the formula describing the planes where these effects occur is derived.

The optical implementation of basic morphologic transforms for greyscale image with a greyscale structuring element is discussed. A novel hybrid linear-and-binary intensity-coding method is proposed and the optical implementation system is described.

We know that optical shadow-casting techniques (OST) may have wide use in optical computing. By using the shadow-casting setup as given in fig. 1, one can get all the 16 logic operations in the central box by (ABCD) of the output screen controlling the switching states of 4 LEDS marked by varies direct as, (beta) , (nu) , (delta) . But the only use of the central box (ABCD) in the output screen and at the same time the nonuse of the other boxes in that screen reflects a shift from the flavor of parallel processings. Here we want to show that the other blocks in the output screen can be successfully used to give us all the 16 logical and arithmetical operations in parallel by creating a proper LED pattern consisting of a minimum number of LEDs and accommodating the tristate number system properly.

The quantization of an image is connected with the introduction of noise. To adapt the quantized image to the demands of its application constraints can be forced on the spectrum of the noise. But not every spectral constraint can be realized. Coupling mechanisms between the values in the image spectrum limit the realization of spectral constraints. In this paper these limitations are examined for different types of spectral constraints. Amplitude and phase control in portions of the spectrum are treated as well as the use of oversampled spectra during the quantization. Halftoning experiments are shown to illustrate the results.

We demonstrate optical two-dimensional wavelet transform in an optical correlator and introduce optical wavelet matched filters that perform the shift invariant wavelet transform for feature enhancement and also the matching with library of templates for pattern recognition with improved discrimination capability and signal-to-noise ratio. Experimental results are shown.