Research On Video Frame Clustering Algorithm Based On Artificial Text

With the rapid development of multimedia,computer communication and Internet, it is simple to store and transport video. Video is widely applied. It is more and more important for the analysis and management of video. How to improve the cluster quality and cluster stability of the video cluster is a hot point in analysis and management of video.In this thesis, firstly, video frame segmentation algorithm of multi-modal information based on artificial text and visual characteristics is proposed. In the video sequence, the video frames containing artificial text are extracted. The video frames are segmented to make use of the text characteristics. The adjacent non-text frames containing artificial text are segmented to make use of visual characteristics. By the way, the segmentation points of video frames are precisely located. Secondly, an improved K-means algorithm is proposed against dependence on the initial value of K in the traditional K-means algorithm. In the improved algorithm, the method of minimum and maximum distances is used to automatically generate the initial number of cluster centers and the initial number of clusters. The dependence on the initial value of K in traditional algorithm is effectively resolved. Meanwhile, in the improved K-means algorithm, it is strict to select the initial cluster centers. The distance of the centers is far. So the situation of the initial cluster centers from the same cluster can be avoided. To some extent, the local optimal state in K-means can be avoided.Results of experiments show that the accuracy of segmentation points of video frames is improved in video frame segmentation algorithm of multi-modal information based on artificial text and visual characteristics. Use improved K-means algorithm to cluster the video frame divided,also improved the accuracy of clustering...