Face Recognition Algorithm Based On Monogenic Binary Coding And Patterns Of Oriented Edge Magnitudes

Face recognition has drawn a lot of attention in recent years and has become the research focus of applied mathematics, statistical learning theory, pattern recognition, computer vision and other areas, since as a biometric identification method, it has advantages against traditional authentication methods.Local feature based face recognition is robust to the changes of expression and illumination. It could achieve state-of-the-art results in large databases such as FERET. Considering one feature alone is difficult to achieve a desired result, in this paper, we propose a feature fusion face recognition algorithm based on local statistics features-- Monogenic Binary Coding (MBC) and Patterns of Oriented Edge Magnitudes (POEM). The main work is as follows:1. We choose MBC and POEM as our face image features. First we describe the related knowledge of MBC, study monogenic signal representation and then we obtain the monogenic representation of face. Different strategies to get the binary coding of the monogenic representation are adopted according to monogenic representation’s physical meanings. At the same time, we adopt local imagery intensity encoding. This encoding method increases the discrimination of the feature and improves the recognition result.2. We introduce a new feature:POEM feature. It uses the LBP operator on the gradient images which are obtained according to gradient orientations. POEM feature is robust to the illumination changes. In this paper, we propose an accelerating computational method of POEM feature and change the process of magnitude accumulation. The method can improve the calculation speed and reduce running time of the system. The experimental results show that the accelerating computational method of POEM feature is faster 50 times than the original method.3. We apply weighted ensemble of piecewise FDA (Weighted EPFDA) algorithm to the obtained MBC maps and POEM map for classification. First we apply PCA and LDA to reduce the dimensions of the features and the redundant information, then we weight each block based on its contribution to the final result by Logistic Regression algorithm. The final similarity is computed with these weights of blocks. Logistic Regression is effective to improve the accuracy of recognition algorithm on FERET sub-databases.We test face recognition algorithm based on MBC feature and POEM feature on FERET database. We achieve 93.77%,90.60%,99.58%,99.49% recognition rate respectively on Dup1, Dup2, Fb and Fc when we do the recognition experiment. For the verification test, we achieve 93.77%,90.60%,99.58%,99.49% on Dupl, Dup2, Fb and Fc when false accepted rate is 0.1%.