Biologically Motivated Multi-Resolution Feature Sensor for ATR and Advanced Object Recognition Purposes

WaveBand Corporation proposes to develop a design and implement a distributed, retina-like biological vision model that performs multi-resolution feature extraction at the sensor level, and outputs transformation coefficients for further processing. The approach incorporates our present understanding of the early stages of the human visual system. It implements a network of complex neurons, novel light adaptation mechanism, and feature extraction processes integrated into a multi-resolution scheme allowing for the subsequent pattern recognition system to capture all important aspects of the input. The system will model low-level feature extraction processes (silicon retina) and combine those in a framework of wavelet-like multi-resolution processing. The Phase I project will develop the necessary algorithms and implement a prototype multi-resolution feature extractor demonstrating the operation and applicability of WaveBand's method. Additional algorithms for automatic target recognition using the sensor will be advanced. The demonstrated system will be used to identify key technology gaps and serve as a testbed for performance evaluation and hardware implementation in Phase II and beyond.