Key-frame extraction based on clustering
Rong Pan, Yumin Tian, Zhong Wang · 2010
The emphasis is on the key-frame extraction technique in content-based video retrieval. Dealing with problems existed in the traditional clustering algorithms, an improved shots key-frame extraction algorithm based on fuzzy C-means clustering is presented. Using the color feature information in the video frames, and then through the improvement of the clustering algorithm of video sequences to acquire the center value of various classes and the membership degree of every frame relative to the classes, finally the shots will be clustered into several sub-shots. According to the relatively uniform of the contents in the sub-shots and the large differences between different classes, as well as the value of the maximum image entropy corresponds to the maximum amount of information in the information theory, the value of maximum entropy frame is extracted as the key-frame from each class. The method overcomes the shortcomings of the traditional key-frame extraction methods that the numbers of the key-frame are fixed. Experiments based on various video sequences show that the algorithm is more reasonable.