Key Techniques Research On Video Retrieval Based On Multi-Features Feedback Fusion Mechanism

With the rapid development of network and multimedia technology, Video information has brought great convenience to people’s lives. However, faced with such a large amount of video data, how to efficiently organization, store, retrieval and browsing these videos becomes a hot research topic in the field of video. The traditional video retrieval methods can’t do better to search the requiring video information because of the richness, diversity, complexity of structure of video data; therefore, the content-based video retrieval technology is becoming a hot research issue. This paper mainly research on the key technologies and methods of the content-based video retrieval, including feature extraction, shot detection, key frame extraction and the video retrieval based on Multi-Features feedback fusion.In the feature extraction, this paper adopts an improved block color feature extraction method which based on equal area of rectangular ring. By means of extracting sub-block accumulative color histogram as color features and setting different weight for different rectangle rings in order to highlight the central part of frame. It shows better effect application in shot detection and key frame extraction.In the shot detection, an improved adaptive shot detection algorithm is proposed. Firstly, the adaptive threshold of detecting abrupt shot and gradual shot is calculated, and the different detection modules is used according to the distance of the characteristics between frames. In the abrupt shots detection, use apart from several frames’frame difference and the edge shape features between adjacent frames to detect the flash. In the gradual shots detection, use the discontinuous between the current frame and back frames to detect the boundary of the gradual shot. The experiment results show that this method has better effect to different types of video.In the key frame extraction, a key frame extraction algorithm based on improved block color features and second extraction is proposed. Initial extracting key frames in accordance with the significant change of frames, then optimizing and selecting key frames in accordance with the frames of location in the video. The experiment results show that the proposed algorithm has good adaptability and the extracted key frames can express the primary content of video effectively.In the Multi-Features feedback fusion video retrieval, based on the improved methods above, this paper adopts a video retrieval method based on color, texture, shape and Multi-Features feedback fusion mechanism. This method uses a single color, texture and shape feature to retrieval video firstly, and then feedbacks grade each result. By many experiments, calculate the different features’ weight used in the process of features fusion according to the single results’mean score. Lastly, compare the results of using single feature and the average weighted feature fusion method. The results show that the recall and precision of this method can reach a better level.