| The automatic recognition of human faces is an active subject in the area of computer vision and pattern recognition over the past few years, which has a wide range of potential applications in the areas of public security, identification of certificate, entrance control and video surveillance. This paper is a study of the recognition of pose varied human faces images. Firstly, human facial cognitive models and rules are studied from the view of cognitive psychology. Then the main algorithms for face recognition are analyzed and compared. A face recognition approach correlative with poses is presented for recognizing pose varied human faces. The presupposition of eigenfaces algorithm is that human faces lie in linear space. However, owing to uncertainty of human faces' poses, recognition rate decreases sharply for the matching of human faces with different poses with the reason that the facial features departures violate the presupposition of eigen faces. The essential idea of the approach correlative with poses is to ensure that face with certain pose is matched in face database with the same pose so that the presupposition of eigen faces will not be violated. Accordingly, corresponding ideas of designing face databases and judging human face' s pose are presented. The preprocessing techniques for human faces' images is presented to reduce the influences of illumination, location and scale variation of faces. An approach of optimizing eigenvectors collection for best describing human faces is presented. 500 faces of 50 persons, incuding all from ORL face database that is used widely by international researcher and some taken by author, are tested in experiments. Eigenfaces gain 88% correct recognition rate, whereas the approach correlative with poses gain 99.6% with two faces refused mistakenly for exaggerated expressions. Experiments results show that the new method gains better recognition rate than eigenfaces, especially for pose varied human faces. |
PDF Full Text Request