An approach to improvise recognition rate from occluded and pose variant faces

Face recognition is increasingly gaining popularity in today´s field mainly because one of the major applications of face recognition, surveillance cameras are being used in real world applications. At the same time, researchers are trying to increase the accuracy of recognition as recognizing face from an unconstrained faces is naturally difficult. In the case of real world application, during image capture there are high chances of faces appearing with different poses, face subjected to illumination and occlusion. In this paper we propose a model that can increase the recognition rate with faces of different pose and faces subjected to occlusion. We introduce the technique of in-painting to restore the occluded face in a frame of video. A dictionary set is created with restored occluded face and faces with varying inclination. In our proposed model, Discrete Curvelet Transform is used to extract features. Comparison with traditional method shows a better recognition rate.