Research On Face Recognition Method For Video Retrieval

With the advancement of science and technology, humans are rapidly entering the information age. The multimedia information represented by videos enriches people's lives and provides security for people's lives; however, on the other hand, how to manage and use these vast amounts of video data poses a challenge to people. Due to the fact that the current videos are mainly characters-based and the development of face recognition technology at home and abroad, implementing video retrieval functions with human faces as the summary information will greatly facilitate management and use of the massive video data for people. Though there are many face recognition methods at present, few methods can meet the requirement of video retrieval because the amount of video data is huge and it requires high calculation speed in such video retrieval. Local Binary Pattern(LBP) based face recognition method can meet the requirement of video retrieval because of its simple and efficient calculation. However, the success rate of traditional LBP based face recognition method dramatically declines when the illumination and background rapidly change. To improve the success rate of face recognition under these changeable conditions, traditional LBP-based face recognition method has been researched and improved in this paper. The simulation results show that the improved method can achieve a higher success rate of face recognition under the condition of rapid changes of illumination and background.The main content of work and general findings in this paper include:1. A face recognition algorithm based on LBP with adaptive threshold is proposed in this paper. This algorithm solves the shortcoming of the original LBP-based face recognition method that it only considers the changes of face details and ignores the face contour information. In this algorithm, the threshold is obtained adaptively by calculating the mean of the absolute differences between neighborhood pixels and center pixel among 3*3-pixel block of an image. It resolves the traditional LBP with threshold's weakness of setting a fixed threshold that leads to dramatically declining of the success rate of face recognition under the condition of the rapid changes of illumination and background.2. A local entropy-based adaptive-weight algorithm for face recognition is proposed in this paper. Considering the fact that different facial organs contribute differently to human face recognition and combining with the amount of information in information theory, the local entropy of each block of face image is calculated and normalized to be taken as the weight of each block. Meanwhile, taking local entropy as the weight of each block reduces the effect of the corrupted sub block caused by variations in illumination and background on recognition result. Simulation results show that this algorithm significantly increases the success rate of face recognition.3. Simulation and comparison of the proposed algorithm are conducted in this paper. Simulation on different face database verifies that the algorithm in this paper increases the success rate of face recognition under the condition of the rapid changes of illumination and background...