A Novel Feature Extraction Technique for Face Recognition

Face recognition has found its extensive application in security. An effective method in extracting features increases the efficiency and the recognition rate of the face recognition system and also makes its implementation easier. This paper proposes a two step methodology for improving the recognition rate of the face recognition system. Face images extracted from an acquisition system posses noise, illumination changes and rotation, reduces the discriminatory power of the classifier. The proposed method involves deriving an illumination insensitive image using Integral Normalized Gradient Image (INGI) and extraction of invariant face features using discrete orthogonal tchebichef moment. Discrete orthogonal moment gives better representation of image even with less order, effective under translation, rotation and tilt and less sensitive to noise. The extracted features are classified using nearest- neighbor classifier. The proposed method is tested using Yale database. Experimental results show the finite number of order for successful feature extraction, the recognition rate under different strategies, the insensitivity of tchebichef moments to noise and the improvement in recognition rate with tchebichef shift invariant. Index Terms-- Face Recognition, Illumination Normalization, Feature Extraction, Tchebichef Moment, Nearest Neighbor Classifier.