Research On Face Feature Extraction And Recognition Algorithms Based On Subspace

Face recognition is one of the important biometric identification technologies. Feature extraction is one of the key problems in face recognition. Many feature extraction methods have been proposed. Among them the subspace methods have been the most popular approach owing to their appealing properties, such as good performance on expression, low time-consuming and validity. This dissertation focuses on the feature extraction technologies based on subspace methods.This thesis pointed out the important significance of feature extraction technology in the science research and industry production based on the background of problem studied, and analyzed the current status and methods of the face recognition. And then the basic principles, features and problems of feature extraction based on subspace methods were summarized. The feature extraction was researched thoroughly and deeply. After making a deep research, we develop three new algorithms as regards feature extraction based on subspace and these algorithms are verified to be effective in the application of face recognition:Firstly, maximum discriminant information projections(MDIP) based on locality preserving projections was proposed. Under the premise of the relative stability of data sets local manifold structure, the algorithm disperses the data points of different classes taking full account of the different face classes, so that it is easier to identify different classes. Under certain conditions, MDIP can be converted into MMC and LPP, and is the organic integration of the two algorithms. Secondly, maximum margin locality preserving projections (MMLPP) based on margin maximum criterion was proposed. MMLPP takes the local and global information of the samples into account by maximizing the between-class scatter, while minimizing the within-class scatter. Effectiveness and stability of algorithm are improved. Finally, uncorrelated locality information projections(ULIP) based on uncorrelated discriminant locality preserving projections was proposed. The algorithm classifies different classes as far as possible while maintaining the local information. The originally far classes are projected far away in the low-dimensional subspace. The discriminative information in the image space was taken full advantage in this method. Besides, a simple uncorrelated constraint is introduced to generate statistically uncorrelated features, which further improves recognition performance.The feasibility and effectiveness of three methods had been demonstrated through extensive experiments conducted on several face databases.