Skip to Main Content
Nowadays, face recognition algorithms, proposed in the literature, reached a correct performance level when the acquisitions conditions for the tested images are controlled (variability of the environment, illumination, poses, expressions and also the number of images in the database to identify people). These performances fall when these conditions are degraded. The controlled conditions of acquisition correspond to a good balance of illumination, as well as a high-resolution, a good pose of the face and a maximum sharpness of the face image. Although several methods have been proposed to resolve the problem of illumination variation, the problem of rotation and occlusion still an obstacle. In this paper, we propose a new method for face detection and normalization which consists to choose the best pose and the point of view of the detected face. This normalization conducts us to select the normalized images to be used as an input in the recognition process, in order to improve its performance. This approach was implemented and tested on a public database. The preliminary results seem very promising.