Research On Key Frame Extraction Algorithm Based On Multi-feature In Video Retrieval

In recent years, with the rapid development of computer network and multimedia technology, and the decreasing cost of storage devices, higher transmission rates, and improved compression techniques, the application range of digital video has been wider. In a broad array of video information, how to get the content quickly and efficiently which users are interested in has become a problem. Therefore, research on video retrieval technology has caused widespread concern.The first step in traditional video retrieval technology is to manually analyze video and label video, and then an index database is constituted by the large number of label information. When users want to query their interested contents, they only need to enter keywords and can find the closest video from the index database.However, there are many problems with this approach. Then the video retrieval technology based on content(CBVR) appears. The CBVR technology can overcome many problems existed in traditional video retrieval method. This article focuses on studying the key frame extraction method in CBVR technology. The key frame refers to the image frame represented the main content of video.This paper firstly introduces the significance of CBVR technology and current research status at home and abroad. Secondly the features and structure of digital video are introduced. Then this paper analyzes the method of segmenting video sequences into one by one small camera shot by detecting gradual and abrupt edge of camera shot. As the core of the study several classic methods of key frame extraction are described. Since the key frames selected by the frame difference method have a problem of large redundancy, Canny edge detection algorithm is used to extract candidate key frames edge. Then redundant key frames are removed according to the edge matching rate of candidate key frames, but this method still has many disadvantages. Based on these existed problem,this paper proposes a new key frame extraction method based on image multi-features. The global features and local features of image are considered, meanwhile color histograms is used to describe the global features of image and SURF feature points is used to describe the local features of image. Firstly the single feature similarity of image is calculated respectively, then the respectively single feature similarity is weighted to get the final similarity, finally the key frames are selected according to the image similarity curve. Experimental results demonstrate that this proposed method is more efficient and the extracted key frame can effectively represent the main content of video.