The Research Of The Face Recognition Base On The Local Feature Analysis

Automatic face recognition (AFR) is a challenging research topic in the field of the pattern recognition. AFR gradually attract more and more extensive concern and high attention from many government organizations and research institutes because of its wide range application in law, security and military affairs. A central issue to a successful approach for the face recognition is how to extract discriminate feature from the facial images.Feature extraction is one of the elementary problems in the area of the pattern recognition. Extracting the discriminate feature is a key aspect of completing the recognition mission. In recent years, local algorithms obtain extensive application in the area of feature extraction due to its insensitive to illumination, change of the facial expression, et al. Under the background of using the face recognition, the thesis carries out a related research on the topic of the Local Feature Extraction algorithm. The main work and contributions of the research can be summarized as follows:(1) The face recognition rate can be influenced by the facial images' scale, rotation and illumination. To address these problems, the thesis introduces into the SIFT (Scale Invariant Feature Transform) algorithm. The thesis proposes an approach based on the specific characteristic of the facial images. Firstly, the thesis carries on key points face image detection with the DOG descriptor. Secondly, the thesis uses Euclidean distance for matching of key points. Thirdly, the thesis eliminates error matching. The experiments show that SIFT method maintain a certain level of invariance to image change factors, such as face image scale and rotation, affine transformation,perspectives change, illumination, noise, and they also show that SIFT maintain the better matching to the image noise and other factors.(2) To address the problems of the robust and efficiency, the thesis presents an improved local LVP histogram algorithm for face modeling and recognition. Firstly, the thesis computes LVPS dictionary by K-means clustering. Secondly, the thesis exploits the computed LVPS to complete the face modeling. The improved method is more efficient than the traditional one for face modeling. Thirdly, the thesis uses weighted histogram of block LVP for recognition. The experiments demonstrate the competence of the proposed algorithm.(3) To address the problems of the High Dimension and robust in the field of the face recognition,the thesis brings up the method of the fusion of the LGBP and SIFT.Firstly, the Gabor operator combines the LBP operator to extract the facial feature. Secondly, the SIFT operator is introduced into the face recognition to extract the facial feature. Thirdly, the thesis extracts the new feature by the feature level fusion of the LGBP and the SIFT feature, and select the 50% feature as the new feature which will be classified. The experiments show that the algorithm not only reduces the dimension but also has higher recognition rate than the signal algorithm.