Research On Key Frames Extraction In Videos

With the rapid development of computer network and multimedia information technology, people's life has been inseparable from the network. Information has been obtained mostly from the Internet. Information resources from the network have been divided into three kinds:video, image and text, and video and image are in a dominant position. Videos are more intuitive and have rich content so that people can through personal experience to get useful information of their own. They can be widely used in various areas of daily life such as education, health care, television, sports and so on. However, videos contain large amount data and complex structure, as a result that it is difficult to access and retrieve videos fast and efficiently. Key frames are the image frames to describe the main contents. The technology of key frame extraction is the key technology for video retrieval, the technology of key frame extraction has been studied in this paper.This paper introduces the research background of video retrieval and key frame extraction technology and research status at home and abroad first. Then some popular key frame extraction methods have been described and analyzed in detail, and each method has been tested. Currently, common key frame extraction algorithms are:1, method based on shot boundary; 2, approach based on the content of images; 3, method based on clustering; 4, method based on motion analysis; 5, approach based on compressed videos. These methods have their own advantages and disadvantages and some limitations, are suitable for specific videos and do not have a wide adaptability. Aimed to these shortcomings of above approaches for key frame extraction, this paper proposes a new key frame extraction method which is based on image information entropy and edge matching rate. The approach firstly calculates the information entropy of each frame, selects corresponding frames whose values acquire local extreme value as candidate key frames, then extracts edges of candidate key frames by Prewitt operator, and matches edges of the adjacent frames. If the edge matching rate of adjacent frames is not less than 50%, the current frame is a redundant frame and discards it. This method does not need to set threshold in advance, dynamically determines the number of key frames according to the content of videos, and has a good adaptability. Various kinds of videos are tested to extract key frames by this approach. Key frames extracted represent the main content of corresponding videos and provide a good foundation for content-based video retrieval and video detection.