Techniques for Indexing MPEG Compressed Videos

The increasing use of multimedia streams nowadays calls for the development of efficient and effective methodologies for manipulating databases storing these streams. Moreover, content-based access to multimedia databases requires effective indexing schemes to facilitate the retrieval of the intended data. This work proposes two techniques to be used in generating color and texture indexes for MPEG video data. At first, the input video data are segmented and key frames are then selected to represent each shot. Afterwards, the first indexing scheme derives color indexes for selected key frames directly from compressed data while the second scheme is used to produce texture indexes to represent the selected key frames. The feature vectors (color and texture) are then stored as metadata for this particular input clip. Experimental results show the effectiveness of the proposed indexing schemes.