Video retrieving technique is widely used to extract videos from thousands of videos available in the database. CBVR is the most suitable technique for fast and accurate retrieval of videos. This is widely used in content linking and brand monitoring. We use an innovative technique where key frame from the input video are used to retrieve those videos having the similar key frame from a large database. The required video can be retrieved from database by matching with the similarities of the key frame given by the user with the similar features of the videos saved in the database. A good color and texture feature extractor is used for extracting the major features from the key frame hence it’s the main module of the CBVR. The main scope of the work is to introduce an improved representation and analysis which helps to retrieve the video from the vast collection of database. Key frame which contain anisotropic and oscillatory textures are very difficult to analyze and represent. This was considered as the major drawback of the existing CBVR system. Hence a new combination of Color Co-occurrence Feature and Wave Atom Feature extractor is proposed in this work. Wavelet, Contourlet and Curvelet are the other multi resolution features. As an advantage, Wave Atom can adapt to any local patterns and can represent those are anisotropic and a novel similarity score method is used for retrieving the video from the database. Two of the widely utilized benchmark databases such as Corel and MITvisTex has been used. From the experimental results, the proposed system outperformed the traditional method and modified both the precision and recall rates for getting better results where the average precision rate was 0.97, recall rates was 0.86 and F-measure was 0.90.
Volume 11 | 04-Special Issue
Pages: 547-556