Research On Illumination Robust Local Binary Face Feature Description Method

With the improvement of people's awareness and requirement about social public security and national defense security constantly, face recognition technology, which is the representative of the biometric recognition technology, is suffered great attention of researchers. Constrained face recognition technology has achieved a satisfactory recognition rate, but the recognition performance of unconstrained face recognition technology will decline substantially, which is suffered from external factors, such as illumination, posture, shade and so on. Therefore, unconstrained face recognition has become the most concern of the current field. Face image description is one of the core characteristics of facial recognition technology. As illumination is one of the key factors, it becomes the hot spot of face recognition naturally. How to extract robust face feature in complex light is the key to solve the illumination problem.Based on the analysis of the methods of face recognition and local binary pattern, the illumination problem of face recognition is chosen as the research focus of this paper. Through research on the methods of local binary pattern and illumination invariant extraction, it is found that local binary pattern has the ability of efficient image texture information description, lacking the robustness of illumination variation. Illumination invariant extraction method has better illumination robustness and worse performance of image texture information extraction. According to the advantages and disadvantages of two methods, the two methods are combined and a illumination robustness combined local partial derivative histogram feature extraction method is put forward. Firstly, the method uses homomrphic filter and relative gradient image to eliminate the influence of illumination factor on image. Then it uses the way of image texture refining to enrich image texture information and solve the problem of the insufficient image texture information. This way also improves the identification and illumination robustness of description feature. Lastly, according to the extraction process of local binary pattern feature, illumination robustness combined local derivatives histogram feature is produced, which has more essential feature of image and neighborhood information. This method enhances the performance of face recognition system effectively.The main work in this paper can be briefly summarized as follows:(1)The present state of the research is introduced in face recognition technology and local binary pattern, which is the theoretical basis for the illumination problem in face recognition. Through in-depth analysis of this problem, the illumination invariant extraction method is chosen. Local binary pattern fusion research plan is the research emphasis in this paper.(2)In view of the basic and related content of local binary pattern and several local binary improved face recognition methods are studied. Firstly, researching basic principle of local binary pattern and it's related contents, it is understood that local binary pattern has fast and efficient image texture description properties. Then, through researching several face local binary algorithms, the improvement thought of each method is analyzed and summarized, which is based on the specific issues of local binary pattern. It provides some ideas to solve the illumination problem. Lastly, several methods of comparison experiments are finished and experiments results are analyzed.(3)Several methods of illumination invariant extraction are discussed and studied. According to the advantages and disadvantages of local binary pattern and illumination invariant extraction, a illumination robustness local binary face feature extraction method is proposed, which fuses the advantages of two methods. Through the experiments finished by FERET, Yale B, PIE face database, the proposed method of illumination robustness and performance recognition is verified.