Face Recognition On Large-scale Database

After more than 40 years' development, great progress in face recognition technology has enabled its commercial application. However, several practical problems keep face recognition far from mature, especially sensitivity to low face recognition rate and long face recognition time on large-scale database.In this paper, we give deep analysis on face recognition on large-scale database, especially focusing on solving the problem of low face recognition rate and long face recognition time.We propose several new algorithms. The main research work and contributions of the paper are as follows:(1) Testing and evaluation of the performance of current standard subspace analysis algorithms on large-scale database. The performance of current standard subspace analysis algorithms on large-scale database is firstly discussed.Based on the theoretic analysis and experimental results in this paper, we conclude the following conceptions which would provide reference signifanance for further research on subspace analysis algorithms on large-scale database: kernel based nonlinear subspace analysis algorithms outperform linear ones on large-scale database. And among all the kernel based nonlinear subspace analysis algorithms, kernel based supervised locality preserving projections algorithm (KSLPP) has the best performance.(2) A feature extraction algorithm which is suitable for large-scale database is proposed—supervised multi-scale kernel based locality preserving projections algorithm. In this paper, firstly, Gabor filter is designed to extract the multi-scale features from the whole face images. Then a two-directional 2DPCA algorithm is utilized to reduce the dimension of the Gabor feature vectors. Finally, kernel based supervised locality preserving projections algorithm (KSLPP) is applied to the resultant feature vectors to extract robust and discriminative features for recognition. Experiments also demonstrate the discrimination power of the new algorithm on large-scale database.(3) Cluster based fast searching algorithm on large-scale face database is proposed. The algorithm firstly does the offline process, including the training to generate a feature space, and then the whole face image database is projected into the feature space to generate face feature database. Next, the improved K-means clustering algorithm is introduced, and the samples in the face feature database are offline clustered to some subsets, and at the same time the image index is generated. Then a new search model is proposed. Experiments show the algorithm greatly enhances retrieval efficiency on large-scale face database.(4) A face recognition system on large-scale database has been primarily implemented. It can automatically detect face, position features of the face, and send it to classifier to recognize. Besides, it realizes functions including management of face image database, face feature database, personal information database et al.