The United States of America As Represented by the Administrator of the

California Institute of Technology

The United States of America As Represented by the Administrator

Hua-Kuang Liu

Tien-Hsin Chao

Li-Jen Cheng Cheng

Chwan-Hwa Wu

Jacob Barhen

Nabil H Farhat

Abdul A Awwal

Shaomin Zhou

Mohammad A Karim

A unipolar terminal-attractor based neural associative memory (TABAM) system with adaptive threshold for perfect convergence is presented. By adaptively setting the threshold values for the dynamic iteration for the unipolar binary neuron states with terminal-attractors for the purpose of reducing the spurious states in a Hopfield neural network for associative memory and using the inner-product approach, perfect convergence and correct retrieval is achieved. Simulation is completed with a small number of stored states (M) and a small number of neurons (N) but a large M/N ratio. An experiment with optical exclusive-OR logic operation using LCTV SLMs shows the feasibility of optoelectronic implementation of the models. A complete inner-product TABAM is implemented using a PC for calculation of adaptive threshold values to achieve a unipolar TABAM (UIT) in the case where there is no crosstalk, and a crosstalk model (CRIT) in the case where crosstalk corrupts the desired state.

Unipolar terminal-attractor based neural associative memory with

A system for optically providing one-to-many irregular interconnections, and strength-adjustable many-to-many irregular interconnections which may be provided with strengths (weights) w.sub.ij using multiple laser beams which address multiple holograms and means for combining the beams modified by the holograms to form multiple interconnections, such as a cross-bar switching network. The optical means for interconnection is based on entering a series of complex computer-generated holograms on an electrically addressed spatial light modulator for real-time reconfigurations, thus providing flexibility for interconnection networks for large-scale practical use. By employing multiple sources and holograms, the number of interconnection patterns achieved is increased greatly.

Reconfigurable optical interconnections via dynamic computer-generated

An optical implementation of an inner-product neural associative memory is realized with a first spatial light modulator for entering an initial two-dimensional N-tuple vector and for entering a thresholded output vector image after each iteration until convergence is reached, and a second spatial light modulator for entering M weighted vectors of inner-product scalars multiplied with each of the M stored vectors, where the inner-product scalars are produced by multiplication of the initial input vector in the first iterative cycle (and thresholded vectors in subsequent iterative cycles) with each of the M stored vectors, and the weighted vectors are produced by multiplication of the scalars with corresponding ones of the stored vectors. A Hughes liquid crystal light valve is used for the dual function of summing the weighted vectors and thresholding the sum vector. The thresholded vector is then entered through the first spatial light modulator for reiteration of the process cycle until convergence is reached.

Optical implementation of inner product neural associative memory

A self amplifying optical pattern recognizer includes a geometric system configuration similar to that of a Vander Lugt holographic matched filter configuration with a photorefractive crystal specifically oriented with respect to the input beams. An extraordinarily polarized, spherically converging object image beam is formed by laser illumination of an input object image and applied through a photorefractive crystal, such as a barium titanite (BaTiO.sub.3) crystal. A volume or thin-film dif

Self-amplified optical pattern recognition system

An inner-product array processor is provided with thresholding of the inner product during each iteration to make more significant the inner product employed in estimating a vector to be used as the input vector for the next iteration. While stored vectors and estimated vectors are represented in bipolar binary (1,-1), only those elements of an initial partial input vector that are believed to be common with those of a stored vector are represented in bipolar binary; the remaining elements of a partial input vector are set to 0. This mode of representation, in which the known elements of a partial input vector are in bipolar binary form and the remaining elements are set equal to 0, is referred to as trinary representation. The initial inner products corresponding to the partial input vector will then be equal to the number of known elements. Inner-product thresholding is applied to accelerate convergence and to avoid convergence to a negative input product.

Imer-product array processor for retrieval of stored images represented

A relatively small and low-cost system is provided for projecting a large and bright television image onto a screen. A minature liquid crystal array is driven by video circuitry to produce a pattern of transparencies in the array corresponding to a television image. Light is directed against the rear surface of the array to illuminate it, while a projection lens lies in front of the array to project the image of the array onto a large screen. Grid lines in the liquid crystal array are eliminated by a spacial filter which comprises a negative of the Fourier transform of the grid.

Large area projection liquid-crystal video display system with inherent

A real-time, pre-detection optical dynamic range compression system uses a photorefractive crystal, such as BaTiO.sub.3 or LiNbO.sub.3, in which light induced scattering from crystal inhomogeneities of the optical input occurs as a nonlinear function of the input intensity. The greater the intensity, the faster random interference gratings are created to scatter the incident light. The unscattered portion of the optical signal is therefore reduced in dynamic range over time. The amount or range of dynamic range compression may be controlled by adjusting the time of application of the unscattered crystal output to the photodetector with regard to the time of application of the optical input to the crystal.

Real time pre-detection dynamic range compression

A method and apparatus are disclosed for real-time optical broadband pseudocolor encoding of an image. A light polarization spatial modulator is utilized to density encode an incident beam of plane-polarized light; the beam is dispersed into its primary colors; the polarizations of the dispersed beams are analyzed to produce positive, negative, and bidirectionally modualated images which are each separately encoded with a primary color; and the primary-color encoded beams are combined to produce a pseudocolored image. In a preferred embodiment the pseudocolor encoding apparatus comprises a compact white-light projection system and a liquid-crystal television spatial light polarization modulator. The invention makes possible the large-screen display of broadband pseudocolor images from real-world input scenes from a television camera.

Real-time pseudocolor density encoding of an image

A method and apparatus for generation of second-rank tensors using a photorefractive crystal to perform the outer-product between two vectors via four-wave mixing, thereby taking 2n input data to a control n.sup.2 output data points. Two orthogonal amplitude modulated coherent vector beams x and y are expanded and then collimated before directing them onto two opposing parallel sides of the photorefractive crystal in exact opposition. A beamsplitter is used to direct a coherent pumping beam onto the crystal at an appropriate angle so as to produce a conjugate beam that is the matrix product of the vector beams x and y, and to separate the resulting conjugate beam that propagates in the exact opposite direction from the pumping beam. The conjugate beam thus separated is the tensor output xy.sup.T.

Method and apparatus for second-rank tensor generation

System for optically recognizing and tracking a plurality of objects within a field of vision. Laser (46) produces a coherent beam (48). Beam splitter (24) splits the beam into object (26) and reference (28) beams. Beam expanders (50) and collimators (52) transform the beams (26, 28) into coherent collimated light beams (26', 28'). A two-dimensional SLM (54), disposed in the object beam (26'), modulates the object beam with optical information as a function of signals from a first camera (16) which develops X and Y signals reflecting the contents of its field of vision. A hololens (38), positioned in the object beam (26') subsequent to the modulator (54), focuses the object beam at a plurality of focal points (42). A planar transparency-forming film (32), disposed with the focal points on an exposable surface, forms a multiple position interference filter (62) upon exposure of the surface and development processing of the film (32). A reflector (53) directing the reference beam (28') onto the film (32), exposes the surface, with images focused by the hololens (38), to form interference patterns on the surface. There is apparatus (16', 64) for sensing and indicating light passage through respective ones of the positions of the filter (62), whereby recognition of objects corresponding to respective ones of the positions of the filter (62) is affected. For tracking, apparatus (64) focuses light passing through the filter (62) onto a matrix of CCD's in a second camera (16') to form a two-dimensional display of the recognized objects.

Real-time optical multiple object recognition and tracking system and

An image difference detection system is described, of the type wherein two created image representations such as transparencies representing the images to be compared lie coplanar, while light passes through the two transparencies and is formed into coincident images at the image plane for comparison. The two transparencies are formed by portions of a polarization rotation spacial light modulator display such as a multi-pixel liquid crystal display or a magneto optical rotation type. In a system where light passing through the two transparencies is polarized in transverse directions to enable the use of a Wollaston prism to bring the images into coincidence, a liquid crystal display can be used which is devoid of polarizing sheets that would interfere with transverse polarizing of the light passing through the two transparencies.

Real-time image difference detection using a polarization rotation

Apparatus is provided which greatly reduces the intensity of bright portions of an image while only moderately reducing the brightness of dimmer portions of the image, to thereby compress the range of light intensities to facilitate detection of the image. The apparatus includes a light detector device formed by a chip of photorefractive material. A two-dimensional array of light beams from an object to be detected, passes through a beam splitter to form two arrays of light beams. The two arrays of light beams are directed at different angles against a surface of the chip of photorefractive material, the two arrays of light beams forming coincident images on the surface. One of the two-dimensional arrays of beams emerging from an opposite surface of the chip has a lower range of intensities, to facilitate detection of the object despite very bright spots on its image. The other array of light beams energing from the chip has a greater range of intensities than the unprocessed image of the object.

Dynamic range compression/expansion of light beams by photorefractive

An optical processor is provided which facilitates selection of any of a variety of patterns or images which are to be compared with a Fourier transform of a template image, wherein the processor can be constructed at low cost. One of the two images that are to be compared is formed by generating video signals representing the image and using those signals to drive a liquid crystal array through which light passes.

Growing community of inventors

Pasadena, CA, United States of America

Hua-Kuang Liu