| With the growing popularity of the Internet, video retrieval and classification applications have been widely applied. The visual content of the video includes color, texture, shape and motion information, and the motion feature is the unique characteristics of the video distinguished from the image. Motion activity is one of the most important motion information, and it describes the motion intensity of the overall video clip. At the same time, motion activity is one of the motion descriptors of the multimedia content in the MPEG-7standard. Video shot is the basic unit of the video processing application. Therefore, this paper focuses on the video shot detection and evaluation of motion activity.First this-paper briefly analyzes the existing video shot detection algorithms. On this basis, combined with the fractal dimension of the-sub-block of the image, we improve the shot detection algorithm based on color histogram, and got better effect on abrupt cuts and gradual cuts. And then use improved the optical flow algorithm to generate the motion vectors between the adjacent videos, and calculate the activity parameters such as the average of the motion vectors, standard deviation, median, and so on. We will take use of these parameters to assess the activity of the videos. According to MPEG-7, we divide the video motion activity into five levels (1-5,1represents the minimum intensity of motion activity,5is the maximum.).We use a multi-class linear classifier to classify the motion activity of video, and the input of classifier are these parameters obtained by the above calculation, the output is the activity level (1-5) of video. At the meantime, to evaluate the algorithm performance, we organize subjective experimental for all videos, and subjective feelings of people acts as evaluation standard. The experiment proved that the evaluation algorithm based on linear classifiers keeps well consistent with subjective feelings, and got the92%correct rate for the video test set. |
PDF Full Text Request