Evaluation and analysis of illumination normalization methods for face recognition

The illumination variation is one of the challenging problems in Face Recognition under complex lighting conditions. Research community has evaluated performance of illumination normalization methods to some extent, yet there is a need to analyze them in depth using performance parameters like False Acceptance Rate, False Rejection Rate, Recognition Rate, Zero False Acceptance Rate, Zero False Rejection Rate, Equal Error Rate, etc. The paper presents performance evaluation and analysis of five illumination invariant methods namely Self Quotient Image, Non-local Means, Adaptive Non Local Means, Wavelet De-noising, and Adaptive Single Scale Retinex on Extended Yale B face database and CMU PIE face database. It is observed that Recognition Rate at Equal Error Rate is quiet acceptable for Self Quotient Image and Adaptive Single Scale Retinex. Also, Adaptive Single-scale Retinex method gives best performance for more complex illumination conditions.