Frequency domain filtering strategies for hybrid optical information processing

Due to the rapid development of spatial light modulators, optical materials and filter design techniques; real time pattern recognition exploiting hybrid optical correlation is increasingly attractive. The spatial light modulator (SLM) enables signal or image patterns to be encoded as amplitude and/or phase modulation patterns across a directed coherent optical beam. It is the vast computational potential of optical information processing that provides the motivation for the design of spatial filters suitable for implementation on currently available SLMs. A good correlation filter should produce a sharp localised correlation peak in the output plane, and be able to achieve this in the presence of noise in the input plane. Thus optimisation is of great importance in optical correlator systems. The lower frequency components produce a broad correlation peak, whereas the higher frequency-band produces a sharp correlation peak that is sensitive to noise. This suggests that a filter with a band-pass characteristic can be made tolerant to noise and also give good localisation of the correlation peak. Here, spatial frequency band tuning and adaptive filtering are developed for this purpose. Spatial frequency selectivity is found to be very important for the design of a spatial filter, a compromise between correlation peak sharpness and noise robustness is sought. Thus, the tuneable photorefractive filter is assessed and difference of Gaussian function filter is developed. For different noise characteristics the spatial filter parameters must be tuned to give optimised performance, this optimisation process depends greatly on the noise and target object spectral characteristics. Adaptive filter design is developed which integrates the phase only filter with the classical matched filter, where a variable amplitude threshold value is set so that, at a particular spatial pixel location, if the amplitude value is greater than the pre-set threshold, only phase information is recorded; otherwise, both the phase and amplitude information are encoded. The development of the synthetic discriminant function filters as distortion tolerant filters was motivated by the sensitivity of the spatial matched filters to distortions in the input image such as in-plane rotations, out-of-plane rotations and scale variations. In applications it is very important that a spatial filter detects the target object from the input scene regardless of its orientation. The design of synthetic discriminant function filters suitable for implementation on commercially available SLM's is an extremely important feature of current research in the area. Therefore, based on the filter synthetic discriminant function (fSDF), a modified filter synthetic discriminant function filter is developed. Via the filter modulation operator Ml the modified fSDF permits advantageous preprocessing of individual training set images that are used in a linear combination to construct the fSDF, which applies a modulation operator M to the synthetic discriminant function. A relaxation algorithm is used to satisfy the equal correlation peaks rule in the correlator output plane. As the filter modulation operators M and M can be given any functional form, the MfSDF design proposed herein is sufficiently general to be described as a unified filter modulation SDF design. By considering the implementation of the modified fSDF on currently available SLM's, the binary phase-only encoded and the multilevel phase and amplitude encoded modified fSDF, which are suitable for the binary mode SLM and the liquid crystal television respectively, are investigated and evaluated. The evaluation is performed to better understand the image distortion range that can be encoded using the modified fSDF filters. The Wiener filter, which has been used extensively for the image restoration and signal processing, is developed for robust optical pattern recognition and classification. The Wiener filter is formulated to incorporate the in-class image (to be detected) and the out-of-class noise image (to be rejected) into a single step filter construction. A Wiener filter-SDF is thus developed and investigated by applying it to vehicle recognition and laser cutting process control. The joint transform correlator (JTC) provides a popular alternative to the Van- derLugt architecture. To improve its performance, a modified fringe-adjusted filter based JTC is introduced and with a multi-object input shown to ameliorate the noise sensitivity of the fringe-adjusted filter based JTC; this provides a solution that overcomes the difficulties encountered with binary JTC techniques. In order to permit the JTC to accommodate a high degree of image distortion, a SDF based modified fringe-adjusted JTC is developed and investigated to illustrate its ability to deal with noisy multi-class, multi-object inputs