A suite of different image processing algorithms can be optically implemented using different filter functions on one common optical architecture (a correlator). This allows one optical system to address all major operations required in general computer vision. These operations, the associated filters, examples of the use of each, and real-time optical hardware are reviewed. For specific image processing applications, only selected operations are necessary; but the same basic architecture suffices.
The use of nonlinear techniques in the Fourier plane of pattern recognition correlators can improve the correlators’ performance in terms of discrimination against objects similar to the target object, correlation peak sharpness, and correlation noise robustness. Additionally, filter designs have been proposed which provide the linear correlator with invariance properties with respect to input signal distortions and rotations. In this paper, we propose simple modifications to presently known distortion invariant correlator filters which enable these filter designs to be used in a nonlinear correlator architecture. These Fourier plane nonlinear filters can be implemented electronically, or may be implemented optically using a nonlinear joint transform correlator. Extensive simulation results are presented which illustrate the performance enhancements that are gained by the unification of nonlinear techniques with these filter designs.
The joint transform binarization routines of frame subtraction and Fourier plane windowing are reviewed. A binary joint transform correlator (BJTC) fingerprint verification system design based on constant false alarm rate (CFAR) processing is proposed. These techniques are used to show that a BJTC based recognition system can be designed to have very good system performance. The tradeoffs between using all or part of the joint power spectrum and between displaying one or several reference prints in the input plane are addressed through simulation. It is shown that both Fourier plane windowing and multiple reference inputs can be used to significantly improve throughput at a slight cost in system performance. A recognition system with a single reference is shown to operate with a CFAR of 1 percent and a probability of false pass of less than 1 percent through a rotation range of ±3.5 degrees. A three reference system with different rotations of the same print as references obtains the same CFAR and similar false pass performance through a range of ±6 degrees.
We describe an encrypted optical memory using double random phase encoding at the input plane and the Fourier plane. This technique allows the images to be stored as independent white complex stationary processes. Experimental results and computer simulations are presented.
Some important concepts of neural networks are similarity, generalization, invariance and training. Some neural networks are supposed to be able to classify objects according to hidden similarities. All of those concepts are put into question by the consideration first put forward by Watenabe that from a purely logical point of view, similarity is a purely arbitrary concept. It can be shown that this implies that the notion of invariance is also arbitrary , that so-called hidden similarities and generalization cannot exist without some external criteria. Such criteria are either implicit in the training algorithms or must be imposed explicitly. This imposes severe limitations on what neural networks can accomplish. However there are some positive implications; neural networks can be designed to classify objects into arbitrary classes. Applications to optical neural networks and examples will be presented.
Original experimental techniques for obtaining the phase-only modulation and the complex amplitude modulation with the commercial twisted nematic liquid crystal televisions (LCTV) are described. The techniques are based on the Jones matrix theoretical analysis. We use the programmable phase-mostly LCTV to optically implement the phase-only matched filters, phase-only circular harmonic filters, phase-only synthetic discriminant functions and the adaptive composite wavelet matched filters for pattern recognition. We also implement the phase-only computer generated holograms, speckle-free kinoforms and optical kinoform encryption using random phase encoding for security applications.
Solid Optics (SO) is 3D optics without air spaces. In some respects it is a 3D version of Integrated Optics (IO), but it shares many features with Conventional Optics (CO) that IO cannot share. After a general comparison of SO, IO, and CO; we note that for many purposes SO is the best choice. We then discuss general approaches, geometries, and figures of merit for SO systems. The final main topic is a review of the two main types of SO system: linear and right-angle.
This paper describes multiple quantum well (MQW) modulators and spatial light modulators (SLMs) which utilize field dependent changes in absorption, index of refraction, and polarization (birefringence). These optical parameters are changed either by band filling in quantum wells or by modifying the excitonic interaction between electrons and holes. Modulation can be achieved by impressing RF/microwave fields across MQWs in a variety of ways including direct biasing, induction by transporting charges in an adjacent channel (by charge-coupled devices, CCDs, or heterostructure acoustic charge transport, HACT), and by launching a surface acoustic wave (SAW) in a region adjoining MQWs. High contrast Fabry-Perot, birefringent, and HACT spatial light modulators are discussed in detail. One-dimensional Bragg SLMs (which use electrooptic or acoustooptic gratings) and Mach-Zehnder interferometers are also summarized.
We review methods of nonvolatile storage in photorefractive materials, namely thermal and electrical fixing, two-lambda readout, two photon recording, and periodic refreshing. For each method, we briefly review the physical processes that are responsible for counteracting photorefractive erasure, and present recent experimental results.
The characteristics of electron trapping optical material are reviewed and equations for modeling the behavior developed. Classical supervised and unsupervised learning algorithms, suitable for optical implementation, are discussed. The principles, optical set up, experimental laboratory results and limitations are described for three optical learning demonstration systems: a supervised Hebbian optical learning associative memory, a supervised Perceptron classifier, and an unsupervised Hebbian optical learning novelty detector. Results show the potential and limitations of electron trapping optical material for optical neural network systems that use optical learning.
Recent advances in both processor technology as well as algorithms have brought the application of correlation filters closer to reality. In addition to object recognition, correlation filters can be used for scene matching applications. In particular, a form of the filters known as unconstrained correlation filters is shown to be optimal from a detection standpoint. Typical applications may include robot vision, autonomous land vehicle, and registering terrain images.