Research On Face Retrieval In Video And Its System Implementation

The continuous improvement of computer performance and widespread popularity of network makes the application of video a great deal of development. More and more records are saved using the format of video. However, how to retrieve the required information in an efficient way from the huge amounts of video-data to improve the work efficiency has become an urgent problem in the modern society. Based on the problem, the technology of content-based video analysis and retrieval is proposed in recent years, and becomes a research focus in the field of multimedia retrieval. Content-based video retrieval involves many branches and applications, one of which is the technology of face retrieval in video. It promotes the development of face recognition and video retrieval.The paper analyzes the related technology for face retrieval in video. AdaBoost algorithm is used to detect the faces from still images and key frames of the video sequences. The proposed method (SVD combined with improved PCA, SPCA+) is used to retrieve the target human face in video. Furthermore, the framework and each module of video-based face retrieval system is designed and implemented in this paper. The main work of the paper includes the followings:(1) This paper summarizes the research status of video-based face retrieval ’and concludes the problems which exist in video-based face retrieval technology.(2) The process of video-based face retrieval is introduced and the framework is proposed. Besides, technology about the video-based face retrieval is analyzed and discussed in a systematical way. As for the face detection module, AdaBoost algorithm is introduced to detect human faces and applied in the development of face detection module. Several experiments are done to verify the validity of the algorithm.(3) A method for face retrieval in video stream based on SPCA+method is presented to solve the disadvantages of the traditional PCA. Besides, SPCA+method is used to develop the face recognition module. Singular value decomposition is adopted to solve the single sample problem, and PCA is improved through local mean and standard deviation in order to solve the problem of illumination effects.(4) A video-based face retrieval system is designed and implemented. At first, the AdaBoost method is used to detect human faces in image and video. After that, SPCA+is used to retrieve the faces in video. Finally, the distance of face features is calculated and the matching work is completed. The faces in video are recognized and the retrieval results are displayed.(5) According to the degree of complexity in video, several typical videos are selected to test the implemented system. Experiments results show SPCA+method can be used to retrieve faces in video, and the system performs better in simple background videos.