Research On LBP-based Feature Space And Its Application For Automatic Face Recognition

In recent years, more and more attention is paid to biometrics especially to automatic face recognition (AFR) because of its potential values in theory and application. A typical AFR system usually contains the following parts: facial image preprocessing, facial feature extraction and classification while feature extraction, which affects the system in a great extent in the selection of the learning algorithm and the classification algorithm, might lead the success of a system.Derived from a general definition of texture in a local neighborhood, the local binary pattern (LBP) texture analysis operator is defined as a gray-scale invariant texture measure. It was first introduced as a complementary measure for local image contrast. Recently, researchers introduce it into the process of facial feature extraction for AFS and achieve a great success. However, it neglects the distribution of the patterns as well as the diversity of the classification objectives, which are of great importance in view point of pattern classification. Corresponding solutions of these limits are proposed in this paper based on a further analysis of LBP.The primary work of this paper can be summarized as follow:(1) A new facial representation named Fractional Order Local Binary Patterns (FOLBP) is proposed. FOLBP rearranges the distribution of the input patterns by a preprocessing step named Fractional Order Transform (FOT), according to different classification objective. Thus, the rearranged patterns are more favorable for classification.(2) By extending the definition of LBP, a new feature space,ε-LBP, which contains the original LBP feature space, is given. In the newly proposed feature space, optimal features for classification in certain sub-space can be achieved by selecting a proper value of the control factor:ε.Experiments carried out on standard face database FERET, ORL, AR and Yale show that the proposed methods are effective and feasible.Based on the research work, a model system for automatic face recognition is implemented under the IDE Visual C++ 6.0. The system consists of real-time face detection, face recognition and information access.