Pattern Theory Extensions for Vector Quantization

Pattern Theory is an engineering theory of algorithm design which provides a robust characterization of all types of patterns. Given a problem description in terms of a truth table and a special measure of pattern-ness, called Decomposed Function Cardinality (DFC), the theory conjectures that the desired solution among a set of candidates is the one that possesses the lowest DFC. The proposed Phase I research program will extend Pattern Theory algorithms to handle more input parameters and multi-level outputs for vector quantization applications. Vector quantization is a block encoding scheme used frequently in digitized speech and image data compression. This effort will examine the feasibility and utility of Pattern Theory generated vector quantization encoders/decoders. Conventional vector quantization performs a distance calculation to accomplish the encoding process, Pattern Theory could replace this with a much faster look-up table approach. Vector quantization look-up tables using Pattern Theory are developed from training vector examples. Training examples are reproduced perfectly but Pattern Theory generated look-up tables have a power of generalization for non-trained examples. This property will be explored for other applications including automatic target recognition and aircraft corrosion categorization under the optional "bridge" task.