The Research Of Smile Recognition Based On Local Feature Analysis

Face expression recognition is a very important research direction of artificial intelligence. In facial expression, smile is the most contagious emotions of all. The study of smile recognition can effectively promote the development of facial expression recognition. Feature extraction is a core technique of smile recognition. The research objects of existing feature extraction algorithms are mainly based on public facial expression databases which obtained under special illumination, posture etc. They are not robust in natural scene because of a variety of adverse factors such as illumination, posture, shade, race, age, etc. Therefore, it is an urgent and tough problem that increasing smile recognition accuracy in real life. In addition, the public expression databases which are available were established to study the seven kinds of human basic expressions. However, as the most important expression, smile has its unique characteristics. The traditional feature extraction algorithms can’t extract the essence feature of smile expression. This paper developed researches on the two main problems mentioned above. Main innovations and work are as follows.1.To solve the problem of constrained scenarios in public expression databases, this paper build a database mainly based on the smile database named GENKI-4K which is obtained from publicly available Internet repositories of personal Web pages. In addition, some representative images of public expression databases are added into the database which can make it more representative of smile face in real life.2.To solve the problem of lack of study in smile feature extraction, this paper studies the importance of lip features in expressing joy and put forward a smile feature extraction algorithm based on the lip area. This method has not only less calculation but also effectively avoid many complex features of global faces which have nothing to do with the smile face recognition. Compared with the global facial features, the lips features is more effective.3.Artificial feature extractions are not extendable. This paper studies an automatic smile feature extraction method based on deep network and designs a deep auto-encoders network which contains four hidden layers. This method avoids the time consumption of artificial feature extraction and depicts more rich inner information in data by transforming feature step by step.4.To solve the problem of ignoring local structure information and learning adverse factors in images actively, this paper put forward a smile algorithm enter traditional features of lip area as the input of deep auto-encoders network which can effectively combines the advantage of both and strengthen the learning ability of deep network. The algorithm can effectively improve the effectiveness and robustness of smile recognition in real scenes and get higher recognition rate in the database of this paper.