The Research Of Method Of Face Analysis Based On Sparse Representation

Global information technology develops in a fast pace. Intelligent living has become an important aspect that people pursue. Face analysis is the core part of many intelligent applications. More and more reserchers pay attention to this area in recent years. Its key technology including: face detection, face tracking, face feature exacting, feature matching, feature classifying and so on. This thesis according to the requirement of the project has a deep research on two aspect of face analysis: face recognition and facial expression recognition. The thesis introduce some related technology in detail and make some improvement at popular algorithms in recent years.Propose a strong real-time video face recognition algorithm and a better facial expression classification algorithm. The main research work including three aspects as following:1 Sparse Representation Classification has a good performance at face recognition applications, but when used in the video process, it can not provide satisfied real-time property. This thesis proposes an algorithm to solve the problem--Muti Frame Sparse Reprentation Classification. The algorithm combines the LBP fature and the original downsample feture in feature exact part, this descriptor has a better describe ability. Besides, Algorithm take strategy that by taking averge feature of multi frames and then do SRC works have a impressive improvement at the real-time property.Thus when dealing with video, MFSRC has a btter recognition rate and faster process speed.2 Studyed the feature exacting part of facial expression recognition, deal with the problem that traditional main face organ texture feature do not have a good ability of classification, This thesis proposes an feature exact method: feature point movement vector combine local texture. This method using AAM model do the feature point location and tracking, then combining feature point’s local LBP texture form the expression feature. Exprement proves that this feature has a better property for facial expression classification.3 SRC do a good job in classification, when introducing into expression classifications has to deal with the problem that large dimension in feature and dictionary is hard to be completed. The thesis uses the change form of SRC algorithm--collaborative representation in facial expression classification. It solve the problem and by doing experiment prove collaborative representation has a better classification ability than traditional SVM in facial expression classifiacation.