Key Technology Research Of Video Shot Segmentation And Key Frame Extraction

The development of Internet and multimedia technology has enriched our life, and boring text information has been gradually replaced by vivid digital videos. Undoubtedly, digital video which combines the text information, audio and visual information is widely used in television, sports, education, medical and other fields. Moreover, the digital video becomes one of the most important methods of people to get information on the Internet. Because the digital video content is rich and has a large amount of complicated data, can we find the most needed video quickly has become a problem to be solved. As Video shot segmentation and key frame extraction is the key step in the video retrieval, we mainly discuss the shot segmentation and key frame extraction technology in this paper, and some results are obtained.1. Image feature extraction. A feature extraction method based on HSV color space and non-uniform block is proposed. We convert RGB color space to HSV color space, and then quantify it for 72-dimensional feature vector. The images are blocked inhomogeneously and given the center piece of the larger weights. This method considers the overall characteristics of the image, and it can highlight the central part of the image.2. A similarity measure algorithm based on the Tsallis entropy is proposed. This algorithm use Tsallis entropy and Jensen formula to constructed Tsallis-Jensen based distance measurement. It can calculate the distance of the 72-dimensional feature vectors directly among the frames.3. A shot segmentation algorithm using the sliding window to optimize the threshold is proposed. Within the sliding window, this algorithm using a higher threshold to detect the abrupt shot change and a low threshold to detect the gradual shot change. To detect the abrupt shot change correctly, we use the interval distance between frames to detect the flash test. This method can avoid mistakes when we detect the abrupt shot change.4. A key frame extraction algorithm based on the optimal hierarchical clustering is proposed. We use the hierarchical clustering to get the initial clustering results, and then use K-Means algorithm to optimize the initial results, extracted frame from the clustering center as a key frame eventually. This algorithm is not sensitive to initial clustering center and clustering number, moreover, the K-Means algorithm have the advantage of fast convergence speed.