A video shot segmentation method based on manifold dimension reduction

In this paper, the manifold dimension reduction algorithm is applied into the video shots segmentation, in which each video frame is mapped from high dimension into low dimension by using manifold learning method to get the compact feature. First, the distance between every two frames is defined and used as a weight factor. Second, a query frame is selected, the sets of the relevant and irrelevant to the query frame are constructed, and another two weight factors are obtained according to the two sets. Then a transform matrix is obtained, through which the low dimensional feature of each frame can be extracted. At last, the Euclidean Distances among low-dimensional features are used for video shot segmentation. Experiment results on different kinds of videos prove the feasibility and reliability of the proposed method.