Face Recognition Based On LBP Statistical Features

Face is an attractive research content in biological genetics, and as a hot spot face image analysis technology has become the contemporary field of computer vision and pattern recognition. It analysis covers the face image detection, face recognition technologies. Face as an important biological feature of identification has a vivid, image, effective and some unique characteristics and advantages, which maintains very broad application prospects in some respects, for example, security, protection, information gathering and Electronic information. With the development of the computer and the information technology, the research and analysis of the face image is more important and has far-reaching significance, which has attracted many researchers.Face recognition is a quite complicated process, in general, including the face feature extraction, face detection and face classification. Face detection can be divided into static image detection and dynamic image detection, face detection, that is, whether a given series of face images or a video, identify the face number and the face specific location. Face classification and recognition is the following work of face detection, that is, in the known face images, identify the specific person who is the face image, with the person's identity, gender, age and a series of information. It can be seen, whether it is face detection or recognition, has important research value and practical significance. Not only is the whole process of face recognition complicated, but also the multifaceted task of research is a difficult study. This thesis focuses on face image analysis process in two aspects.Experiments show that the random graph theory comparing CCCD (Class Cover Catch Digraphs) classifier with classifier based on nearest neighbor graph has more advantages, and more relatively stable. In recent years, some people have used the classifier that is be based on the boosted CCCD in the field of face detection to achieve a more better results. Therefore this thesie try and research it. In addition, whether face detection and classification, if a better face feature data are selected and effective method of face feature extraction is a prerequisite to ensure good results. The LBP (Local binary pattern) operator has been proved to effectively extract the texture features of face images; therefore, this thesis presents a way to improve theε-LBP operator and CCCD classifier combination for face detection. This operator withε-LBP texture features extracted from face image, corresponding to different values of the extracted face feature data weighted fusion, it is weighted data fusion to describe the features of the face, then the feature data is used for the establishment of CCCD classifier, and effective detection of human face images is achieved.In face recognition, by using high performance of LBP operator and ISOMAP(ISOmetric feature MAPping) reducing high dimensional data, this thesis presents an improved algorithm of combining LBP-based operator with ISOMAP for face image recognition. Face texture feature data extracted by LBP operator need to be reduced with essential characteristics of the data ISOMAP can implement the function. Therefore, firstly, an improved operatorε-LBP is used to texture feature extraction of face images, and ISOMAP data dimension reduction can acquire the essence of geometry from the texture data. Finally, the dimensionality reduction of data are taken as the input for face classification. The experimental results show that, the algorithm can be run well on face image classification and recognition comparing with other algorithms.