The Application Of Binary Feature In Face Recognition

Face recognition problems is one of the hot research topic in the area of computer vision. The performance of binary feature in face recognition is discussed in this paper. The contributions of this paper are as follows:1. The performance of the classic binary features is reviewed, we divide those existing binary features into two categories according to the different extracting methods. The advantages and disadvantages of them are analyzed.2. We propose a new feature named Principal Learning-based BRIEF (PLEB). Each bit of the feature is derived by comparing pixel’s gray value, then PCA transform is used to remove the redundant information. Moreover, the processes of coding and pooling are proposed to improve robustness. Experiments results on the Labeled Faces in the Wild (LFW) database and our own face datasets show that PLEB feature can effectively improve the recognition performance.3. Dimension reduction and null-space are two classical method to solve the small sample problem. The method based on feature reduction is mainly due to the process of whitening process, while the null-space method mainly due to the null space of within class scatter matrix. These two algorithms on the LFW are compared and the experimental results show that the method based on null-space has better performance.4. A webpage-based face recognition demo system is realized, which has the functions of the face picture matching and similar face picture search.