Face Recognition under Lighting Variation Conditions Using Tan-Triggs Method and Local Intensity Area Descriptor

Lighting variation is a specific and difficult case of face recognition. A good combination of an illumination preprocessing method and a local descriptor, face recognition system can considerably improve prediction performance. Recently, a new descriptor, named local intensity area descriptor (LIAD), has been introduced for face recognition in ideal and noise conditions. It has been proven to be insensitive to ideal and noise images and has low histogram dimensionality. However, it is not robust against illumination changes. To overcome this problem, in this paper, we propose an approach using an illumination normalization method developed by authors Tan and Triggs to normalize face images before encoding the processed images based on LIAD. The recognition was performed by a nearest-neighbor classifier with chi-square statistic as the dissimilarity measurement. Experimental results, conducted on FERET database, confirmed that our proposed approach performs better than traditional LIAD method and local binary patterns, local directional pattern, local phase quantization, and local ternary patterns using the same approach with respect to illumination variation.