Illumination and pose Invariant Face Recognition

Abstract

Face recognition system is a kind of biometric system which may or may not require the co-operation of the participant. Whereas the other biometric systems such as fingerprint recognition, retina or iris based recognition need the active co-operation of the participant. Being an example of pattern recognition system, all the steps involved in pattern recognition system are equally applicable in face recognition system. Normally all the face analysis methods are classified into two types- face verification and face recognition. In the verification mode the claimed identity of a person is verified. Whereas in a recognition mode, the person s identity is matched with the one in the database. newlineOver the last three decades number of algorithms are suggested by various researchers for face recognition. The researchers from the diverse fields like computer science, neuroscience and psychology have a huge interest in it. Active research is going on in face recognition and a notable progress is achieved in various phases of face recognition like segmentation, feature extraction and classification. But the robust face recognition is still to be achieved. The multiple evaluation tests and grand challenges like FERET, FVRT and FRGC pointed out that the illumination and the pose variation are the main challenges in the successful implementation of a face recognition system. newlineWe propose a face recognition system based on Gabor wavelet, as it proved the best in the feature extraction. Gabor wavelet bank contains a set of narrowband filters with different bandwidth. With the help of these filters extract the features present in the image and then use these features for further processing. The Gabor features extracted have high dimensionality. For the dimensionality reduction subspace analysis techniques using kernel linear discriminant analysis is used. The use of the Gaussian kernel for projecting the features into a high valued features to make them easy to discriminate with linear discriminant analysis. To reduce the dimensionalit