The Research For Face Recognition Based On Pseudo-Zernike Moment And BP Network

Based on previous study, a novel face image feature extraction based on normalization of pseudo-Zernike moment was proposed in this paper. Firstly ,the pseudo-Zernike moments of the face image were computed .Secondly ,normalized the result .At last ,the nearest neighborhood algorithm was used to construct classes .Experimental results demonstrate the presented approach basically eliminates the effects of the change of face scale and rotation ,and has better robust to variations of illumination ,pose and facial expression .On the basis of Back Propagation and Region Reproduction Algorithm, combined methods of normalization of pseudo-Zernike, a novel face recognition was proposed in this paper. Firstly, training images are extracted by pseudo-Zernike, then normalized by geometrical moment with (0,0). Second, BP network is inputted by the results and trained with Region Reproduction Algorithm, as a result it can fastly converge and avoid local optima. Experimental results show this method is robust against the change in facial expression , illumination and pose, and has higher recognition rate.