FUZZY MEASURE BASED ADAPTIVE METHODS FOR ILLUMINATION INVARIANT FACE RECOGNITION

Abstract

Varying illumination is one of the well known and challenging problems in Face Recognition applications. Numerous methods have been proposed by researchers, but recognition performance under complex illumination is not yet satisfactory. The paper presents Fuzzy based methods to adaptively normalize illumination in face images for Face Recognition under varying illumination conditions. The paper has main two contributions: (1) Fuzzy measure based Adaptive Single-scale Retinex and (2) Fuzzy measure based Adaptive Single-scale Self Quotient Image method. Also, two more variations of these methods are presented. There are two main advantages of these methods, as compared to multi-scale Retinex and Self Quotient methods. Firstly, due to the adaptive nature of proposed methods, discontinuity in facial feature is smoothed and discontinuity due to shadows is preserved and hence performance is better. Secondly, computational complexity is reduced because of single scale 3∗3 filter instead of multi-scale filters. Rigorous experiments have been performed on CMU PIE face database and Extended Yale B face database. For determining False Acceptance Rate, 529 and 550 imposter faces are used for experiments on PIE and Yale databases respectively. Proposed methods are compared with existing methods under same experimental setup using six performance evaluation parameters. Results have shown that Fuzzy measure based methods performs well.