Study On Face Recognition Methods Based On Face Characteristic

The research on the methods of face recognition is in great demand in the pattern recogniton all the time presently. Because face is not the quality of being rigid, it is easily influenced by brow, gesture and lighteness that bring about some difficulties in face recognition. The methods of face recognition studied at present are in the given circumstance. But people ignore the face recognition of the special persons who have the lentigines or the birthmark in their face. These features in the face are called dye spot and dye birthmark in the iatrology. In this text, these features are called by a joint name: face charateristic. Researching on a efficient face recognition method about face characteristic is significant in access control and looking for a person in police. In this paper, from the process of face recogniton, a appropriate method in these special person is made an effort to be found.In the phase of image processing, firstly, the size and gray must be equal. Secondly, the interested area is strengthened by the extending gray. Finally, face characteristic is partitioned in the method of threshold or area-growth.In the phase of feature extraction, according to the transcendent knowledge of face characteristic, the thress methods that are shape feature, local PCA, wave transform are studied. We compare with these three methods. It is known that they have common ground in keeping down the main feature of image and decreasing the dimension. The shape feature method is robust in face gesture. But it ignores the other main information in the face image and is only fit to the rough classification, local PCA is uncertainty in division. Furthermore, it is complex in calculating. Wavelet transform can extract the local information from image. Furthermore, it is robust in lightness and simple in implement. But this method has to decreasing dimension. Thinking over these three methods, this text uses wavelet transform in expressing face image. Base on wavelet transform, face characterisic is used identifiable pixels and extracted gabor feature.In the phase of classification, we compare with the three methods that are nearest-centroid, BP network and LDA. Because LDA not only can embody the diversity within face image, but also can minimize diversity in the same sample matrixs and maximize diversity in different sort of sample. This paper use Gabor wavelet and LDA in face recognition. The last vector of classification combines local Gabor-LDA feature with the whole LDA feature. Moreover, a method of feature fusion based on complex vector is improved. By experimenting, this method can effectively identify face image based on face characteristic.