MULTI-SENSORY OPTO-ELECTRONIC FEATURE EXTRACTION NEURAL ASSOCIATIVE RETRIEVER

Optical pattern recognition and neural associative memory are important research topics for optical computing. Optical techniques, in particular, those based on holographic principle, are useful for associative memory because of its massive parallelism and high information throughput. The objective of this chapter is to discuss system issues including the design and fabrication of a multi-sensory opto-electronic feature extraction neural associative retriever (MOFENAR). The innovation of the approach is that images and/or 2-D data vectors from a multiple number of sensors may be used as input via an electrically addressed spatial light modulator (SLM) and hence processing can be accomplished in parallel with high throughput. A set of Fourier transforms of reference inputs can be selectively recorded in the hologram. Unknown image/data can then be applied to the MOFENAR for recognition. When convergence is reached after iterations, the output can either be displayed or used for post-processing computations. We included experimental results that demonstrate the ability of the system to recognize and/or restore input images.