Research Of Shot Detection And Key Frame Extraction Of Content-based Video Retrieval

With the rapid development of modern multimedia technology and Internet technology,the application of digital video in daily life is becoming more and more common. Relevantresearchers gradually began to focus on how to retrieve and manage these multifariousvideo data fast and efficiently, the research of content-based video retrieval technologyspring up rapidly. These technologies are the indispensable part of a completecontent-based video retrieval system, such as shot segmentation technology, key frameextraction technology, feature extraction and matching, query and retrieval mechanism.Each technique application algorithm will directly affect the performance and efficiency ofdigital video retrieval.Shot segmentation technology and key frame extraction technology are the two keytechnologies of content-based video retrieval system. In this paper, combine these twotechnologies as the key research object, the main research content is as follows:(1)Firstly, the existing various kinds of shot segmentation techniques are analyzed indetail. Thus the dual threshold shot boundary detection algorithm is improved. By usingthis method, the differences between the adjacent frames and the segmentation thresholdhave been determined with the internal mobility of each shot and leaky, at the same time,leaky and false detection can be avoid effectively. At last, substantial experiments provethat it has a high precision rate and recall rate by using the improved shot boundarydetection algorithm.(2) A new method of key frame extraction method based on the color histogram isproposed in this chapter. This method combines the block color histogram and the globalcolor histogram, extracting video frame details comes first by the using of the block colorhistogram method, suspected key frame sequence has been got, and then further extractionproceeds by the using of the global color histogram, the final key frame sequence has beengot. Experiments show that this key-frame extraction method can combine the advantagesof these two methods very well and can be a better representative of video content. It bothcan better describe the content of the lens and will contain too much redundant information.(3)By researching the existing key frame extraction technology, an extraction methodbased on color and contour features integration was proposed. In this method, featureextraction comes first by integrating features of HSV color space and contour features ofHu invariant moments within image block. Thus after video sequences have beensegmented using dual threshold shot boundary detection algorithm, and the threshold hasbeen determined with the differences between the adjacent frames, dynamic clusters are achieved and video key frames are extracted based on segmentation results. Theexperimental results on different types of videos show that the proposed key-frameextraction method can be a better representative of video content, and conductive to therealization of video analysis and retrieval.