09 - Quantization and classification
Gray · Traitement du signal · 1998
The problem of quantizer design for detection or classification has a long history, with classical contributions by Kassam , Poor, Picinbono, Bucklew and others . The goal was to design a quantizer such that a detection rule based on the quantize d information was optimized . During recent years an alternative approach has been developed which seeks to jointly optimiz e quantization and classification by incorporating the Bayes risk resulting from the quantizer into the quantizer optimization . In this paper the general classical approach of Picinbono and Duvaut is compared contrasted with the joint approach an d illustrated by a simple example .