Shift-invariant real-time high-speed optical image correlation system

We have studied, theoretically as well as experimentally, diffraction from volume holographic gratings, and its application to design of optical image correlation system. We have investigated a holographic system for image recognition in real time by implementing a time-efficient exhaustive search in a very large image database with a shift-invariant correlator. Our system has the potential of achieving correlation data processing rate of 10 12 bps. The database of filter images contains multiple copies of the same image to account for rotation and scale variance, while shift invariance is accomplished by using a thin correlator material. The filter images are to be stored in the ultra-high density spatio-angular multiplexed holographic memory with nearly 2TB storage capacity. Real-time recognition of the stream of the incoming query images is performed with a photorefractive holographic correlator. Time-efficient search is accomplished if one uses a real-time VanderLugt correlator, because this architecture allows to overcome the speed limit of the photorefractive gratings formation. For the preliminary experiments, we constructed edge-enhanced VanderLugt-type spatial filters using samples of 6mum thick silver-halide emulsion layer deposited on a glass surface. These samples possessed the shift-invariance properties; however because these materials produced permanent holographic recordings, they were not suitable for real-time operation. We identified photorefractive polymers that can be made as thin as 16mum, and as a consequence posses a large degree of shift-invariance, to be the best materials for real-time correlation. We performed a series of experiments to measure the angular bandwidth of these materials in the phase conjugation process by a degenerate four-wave mixing arrangement. We also demonstrated shift-invariant correlation in the real-time VanderLugt architecture using this material. In addition, we investigated theoretically and experimentally the direction of the diffracted beam for off-Bragg incidence. This analysis is important in designing an actual working optical correlator system, since one must be able to predict the direction of the correlation beam. The work performed under this thesis project has thus established the feasibility of a real-time, high-speed, translation-invariant image identification system. Future work would center on engineering optimization of this artificial vision technology.