Development and performance analysis of human face recognition techniques

Abstract

The main aim of the present research has been to develop face descriptors that are resilient to variations in pose and illumination to some reasonable extent. An effort has also been made to thoroughly analyze and explore the most recent and well-established face recognition algorithms for their fusion to optimize them for practical situations. The present work develops an improved and optimized feature selection approach verified through statistical analysis to find the most discriminative coefficients for the various feature sets. It focuses on minimizing the within-class variations while maintaining the between-class differences, by selecting those coefficients which deliver well-separated class averages. For its validation, it has been applied to select optimized ZM coefficients, named as MDZM, and to improve the PCET technique to DPCET. newline