The Research Of The Key Frame Extraction Algorithm Of Content-based Video Retrieval

With the rapid development of the network technology and the multimedia technology, the vast amounts of video influx into Internet, giving rise the people to the visual and auditory pleasure. Meanwhile, on the massive scientific and rational management of video has become a difficult task before the people. How people find video datas that they are intersted in in the flood of video has become a problem to be solved by the researchers of computer technology. Content-based video retrieval technology become one of hot issues of computer research in this background.Content-based video retrieval by extracting the color, texture, shape, motion and other visual characteristics of video content as an index, provide users with a feature-based visual content similarity search tools, according to video content and context of the relationship ,search the video data in the large-scale video database, the system does not require human intervention, automatically extract and describe the characteristics and content of video. Content-based video retrieval technology combines image processing, pattern recognition, computer vision, image understanding, cognitive science, artificial intelligence, database management systems, knowledge systems, human-computer interaction and other areas of science and technology.Content-based video retrieval key technologies include: shot detection, key frame extraction, shot clustering, video data organization and indexing, feature extraction and matching, interface display and interpersonal interaction, and evaluation standards.Both of video shot detection and key frame extraction are the important parts and crucial steps of content-based video retrieval system, related to the video summary generation, the establishment of video database indexing and retrieval efficiency.Video shot detection is the key-frame extraction of the prerequisite and basis , the efficiency of video shot detection algorithm is a direct impact on the results of key-frame extraction.Video shot detection algorithm can be divided into video cut detection algorithm and video gradual shot detection algorithm according to two types of shots. For the shot cut detection, shot cut detection algorithm based on HSV color space of video is proposed, which frame image by calculating the mean and variance of the HSV color space to the difference degrees between the adjacent two frames, according to the size of the difference degrees to determine the position of the shots convertment. For the shots gradient detection , video shot gradient detection algorithm is proposed based on the difference between frames, by analying the curve of the difference between the two adjacents. The problem is proposed that the similar of the fluctuations of the shots gradual process and factors of movement, and the process by calculating the fluctuation of the difference before and after the two to determine that the fluctuation is caused by the shots gradual or movement of factors.After video shot detection,complete video material is divided into many a shot, every shot is that variable number of several frames form a set. Several frames in the same shot are great similarity, resulting in considerable redundancy. To solve this problem, the key frames must be extracted in each shot as the representative frames. In the research process of the key frame extraction algorithm, to face the issue that the key frames'judging guidelines are very ambiguous.In this paper, the key frames'judging guidelines have been proposed.Constraints of these guidelines, based on the block of frame image of key frame extraction algorithm is proposed . The main idea of the algorithm is cutting the key frame into several sub-blocks, according to calculating the characteristics difference of two sub-blocksthat are the correspond-ing position. The partially dramatic changes in sub-block will be as a key frame extracting. The algorithm for capturing the main movements or a significant change in the partial frame images in the same shots is proved to be effective. In order to improve the efficiency of key-frame extraction, and taking into account the frames'continuity in the same shots, the idea of the binary search method is introduced into key frame extraction, reducing the key-frame extraction algorithm's time complexity.This experimental video datas used are from the Internet, selected as the experimental material of video datas have certain characteristics. The experimental results show that the proposed algorithm for video shot detection algorithm is more accurate,and with having strong robustness. The proposed key frame extraction algorithm for key frame extraction has higher extraction efficiency,the key frames can reflect the content of the shots effectively.