Abstract
— In this paper a novel technique of content based video retrieval is presented. The proposed technique uses Dual Tree Complex Wavelet Transform (DTCWT) based features of video frames for the purpose of shot change detection, key frame selection and video indexing. For shot change detection consecutive frame difference is computed, shot change is reported when the difference exceeds a certain threshold. For keyframe selection a frame is to be selected which is not part of shot transition using k-mean clustering of DTCWT feature vectors. Video shots are indexed using DTCWT features of the selected keyframes. Video query is processed by comparing the features of shot with the features database of the shots. For the purpose of features similarity we have used correlation based distance metric as it produced better results for this kind of feature similarity. The results are compared the results with classical techniques and it is shown how dual tree complex wavelet transform based features performed better. The whole framework uses similar kind of feature which makes it simple and efficient.— CBVR, Video Indexing, Shot Boundaries, Key Frames