关键词: 计算机应用, 人脸识别, 区域性-方向性加权的二元局部纹理, 多尺度小波分解, 直方图特征

Abstract:

In order to improve the recognition rate and robustness of face recognition algorithms, the impact of different directional details of facial images on recognition rate is analyzed, and a Horizontal component prior principle (HCPP) is proposed. Then a robust face recognition algorithm based on Regional directional weighted local binary pattern (RDW-LBP) is proposed, in which HCPP is employed. The face image is decomposed by multiscale Haar wavelet to extract multi-level scale coefficients and the first level horizontal detail coefficients, which contain the most effective details are used for recognition. These coefficients are regarded as the elements of matrix M to be analyzed. Then matrix M is divided into several regions, from which the RDW-LBP feature distributions are extracted and concatenated into an enhanced feature vector for use as a facial descriptor. Finally, the classifier based on Chi-Square distance is used for recognition. The correctness and validity of HCPP are verified by test results. Compared with LBP descriptor proposed by Ojala, RDW-LBP descriptor improves the recognition rate efficiently without increasing computational complexity.

Key words: computer application, face recognition, RDW-LBP, multi-scale wavelet decomposition, histogram features