Video Shot Retrieval Using Multi-feature Approach

Abstract

The evolution of Internet and mobile technologies has raised the growth of video data that spur demand for efficient browsing and retrieval technologies. Content-Based Video Retrieval (CBVR) is evolved progressively to retrieve desired videos proficiently from large repositories based on the video content. In this aspect, an efficient CBVR method is presented using multiple features from video shot keyframes. The approach constructs histogram for Sobel magnitude of V-channel from HSV image and also local binary cumulative sum variance pattern for grayscale image. Histograms, thus, constructed are concatenated to build multiple feature vector database. Further, shot matching process is established by applying Euclidean distance between shot keyframe and query keyframe features. Experimentation was carried out on UCF YouTube action benchmark dataset to analyze the efficiency of the proposed algorithm. Video shot retrieved results depict significant improvement of the presented CBVR approach in comparison with baseline algorithm in terms of evaluation metric.