Advanced feature set for face recognition under assorted lighting conditions

Matching face image under assorted lighting conditions is a challenging task in a face recognition system (FRS). The feature extraction method used in Local Binary Pattern (LBP) fails in detecting the eye position and it is less sensitive to noise in uniform image regions. This paper proposes an Advanced Ternary Pattern (ATP) feature set to overcome these issues. The issue on noise sensitivity in uniform image regions such as cheeks and forehead is attempted to resolve using the normalization technique which regularizes all the extracted features under various lighting conditions. ATP feature set uses the Efficient Feature Orientation (EFO) method for effective and accurate face normalization. The proposed model improves the Face Verification Rate (FVR) up to 90% for the less dark images and about 78% for fully dark images. Experimental results show that the proposed scheme makes FRS insensitive to illumination.