Video Face Clustering Via Constrainted Sparse Representation

With the development of information technology, a large amount of video data have been produced. The video data contains more information compared with text and image in terms of the continuity on time. Therefore, the video processing techniques become more attention. In addition, extracting useful information from video is still a challenge task since the large amount of video data. And it has far-reaching theoretical significance.As an important branch of video process technique, video face clustering is a wide-used technique in many applications, such as, content-based retrieval, automatic cast listing in feature-length films, rapid browsing and organization of video collections and so on. Traditional video face clustering methods always be developed based on image clustering techniques which just use the information of vision of face images. In real-world videos, lighting conditions, facial expressions and head pose may drastically change the appearance of faces. Partial occlusions caused by objects such as glasses and hair also cause problems. As a matter of fact, the video provides some beneficial inherent advantages that can be used in clustering to improve the clustering performance. Concretely, two types of inherent pairwise constraints can be easily extracted from a video that is inner-track and inter-track relations which is regarded as must-link and cannot-link constraints respectively. Specially, the must-link constraints come from a face track and the cannot-link constraints come from the overlapped face tracks which means that there are faces occurring in a video frame from different face tracks. These constraints are further effectively incorporated into our novel algorithm, Video Face Clustering via Constrained Sparse Representation(CS-VFC). The CS-VFC utilizes the constraints in two stages, including sparse representation and spectral clustering. In the stage of representation, the constraints are utilized for exploring the unknown relationship among faces. In the stage of spectral clustering, a soft manner is taken for maintaining the structure of relationship of faces and farthest satisfying the constraints simultaneously.To verify the efficiency of the proposed method, three real-world video datasets have been used in our experiments comparing with several state-of-the-art-methods on several indice. The results have shown the efficiency and feasibility of the proposed method.