A Research Of Face Feature Extraction And Recognition Algorithms

Human face recognition is an important research subject in the field of biometric recognition nowadays, which involves the science of computer vision, pattern recognition and image processing. The face recognition techniques have been widely used in many fields of society security and information safety, the research of which is of high academic value and practical significance. The human face features can be affected by the environmental factors and the object variations, which leads to the research of face recognition as a very challenging task. The key issue in the face recognition techniques is to extract the efficient and accurate face features under the uncontrolled environment. In recent years, many excellent algorithms of face recognition emerge in endlessly, among which the methods based on the local texture analysis have the prominent effect on human face images. Such algorithms based on local descriptors always have advantage in the local feature representation and show the robustness to the noise interference.In view of the superior characteristics of the local analysis methods against the unfavorable factors, this paper focuses on the local pattern analysis and its application in the feature extraction and recognition of human face images. The main research works are as follows:1. The local pattern theory is investigated in this paper. The local pattern theory based on the local image regions has accomplished the overall consideration of the local feature and global information in the human face images by the pattern representation and synthesis analysis, which can be used to solved such the highly information-coupling problems as the human face recognition. The successful applications of local binary patterns, local ternary patterns and other local patterns in the field of the human face recognition have shown that the local pattern analysis is feasible and effective for the human face feature extraction and recognition. Among the local pattern analysis, the selection of feature basis, the solution of pattern coding and the methods of pattern synthesis are three key issues which are crucial to the success of the recognition algorithms.2. The Local intra-directional pattern (LIrDP) and local inter-directional pattern (LIeDP) are proposed and studied in this paper. Both of these local descriptors are based on the Sobel operators. However, they have the significant differences in the nature and concrete operations. LIrDP mainly focuses on the relative variations of image pixels in local region under the same condition. The LIeDP features are extracted on the basis of the different responding value of the local region to the various directional gradient operators. The recognition algorithms based the LIrDP and LIeDP are proposed in this paper separately, which are used to evaluate the recognition performance of the LIrDP and LIeDP on FERET database and Extended Yale-B database. The experimental results prove that the recognition algorithms based on the LIrDP and LIeDP outperform the available local pattern algorithms.3. The face recognition algorithm based on the complete local directional patterns is proposed and investigated in this paper. The complete local directional patterns have realized the integration of the two kinds of local features from the LIrDP and LIeDP, which provide more comprehensive analysis for the local face features. The face recognition algorithms based on the complete local directional patterns are introduced in this paper and the experimental results have demonstrated its effectiveness.4. The structural diffusion relations (SDR) and their application for human face recognition are proposed and explored in this paper. The structural diffusion relations are based on the micro-structures of the local image regions. In the local pattern analysis, the micro-structures are represented via the numerical values computed by the various local descriptors. The basic contents of the structural diffusion relations and their application in the field of human face recognition are provided in this paper. The experimental results of recognition algorithms based the SDR and local descriptors have illustrated that SDR can provide more contributions to the recognition rates.