FACE RECOGNITION ALGORITHM BASED ON OPTIMAL WEIGHTED MULTI-DIRECTIONAL LOG-GABOR FEATURES

Abstract

In order to improve the performance of face recognition, a face recognition method based on an adaptive fusion of multi-directional and multi-scale Log-Gabor features is proposed. First, the multi-direction and multi-scale Log-Gabor transform is applied to face images. Then the Log-Gabor feature maps of different scales are fused to extract the Gist vectors of the fused feature maps. In addition, the mutual information algorithm is used adaptively to calculate the weights of Gist vectors. Finally, the adaptive weighted Gist vector is used as the feature vector of face images, and the K-nearest neighbor classification algorithm is used to realize face recognition. According to the face recognition experiments on WIDER FACE data set and MAFA occlusion face data set, the accuracy of the proposed method is presented. Compared with the classical face recognition algorithms, the effectiveness of the proposed method is verified.