Research On Shot Segmentation And Key Frame Extraction Method In Video Retrieval

With the advent of the Internet+ era, multimedia is facing new opportunities and challenges in the new situation, but also bringing new thinking of how to integrate multimedia industry with Internet thinking. The rapid growth of multimedia information also makes the development of multimedia technology particularly important, how to effectively organize and manage video information, and quickly retrieve required video from video data has become the research focus of majority of researchers. Content based video retrieval technology arises at the historic moment under such background, which can analyze and process the content of the video, so as to realize the retrieval of video information. Video shot boundary detection and key frame extraction are two key techniques in the content based video retrieval, and it is also the basis for the realization of the function of video retrieval. Therefore, the main research of this paper is shot boundary detection and key frame extraction.The shot boundary detection algorithm based on the combination of HSV color histogram and HOG features is presented, which includes two algorithms, including abrupt shot detection and gradual shot detection. In the process of abrupt shot detection process, HSV color histogram is first used to detect the position of the abrupt lens. Because HSV color histogram method can not represent the spatial position information and will be affected by the brightness transformation and so on, HOG feature is then used for second detection in order to improving abrupt shot algorithm in precision and recall. In the process of gradual shot boundary detection, HOG feature and HSV color histogram feature are extracted by 25 frames and 5 frame intervals respectively, and then comprehensibly used for gradual detection, which can accelerate the detection speed of gradual shot detection. Moreover, the ideal detection results are obtained.In key frame extraction, on the basis of research and analysis on the key frame extraction algorithm, the key frame extraction algorithm based on 2DPCA and density peak clustering algorithm is presented. 2DPCA dimension reduction method is firstly used for dimension reduction of image feature, which makes image data less redundant; moreover, the detection time is saved. The density peak clustering analysis method is then used to improve drawback of the traditional algorithm in the number of key frame extraction which is not flexible enough and directly excludes the image frames affected by flash. The experimental results show that the presented key frame extraction algorithm can accurately locate the key frames, and the extracted key frames are representative from two aspects of subjective and objective point of view.