Midstream Content Access of Visual Pattern Coded Imagery

Multimedia retrieval is going to play an increasingly important role in the future. Since the availability of digital media seems ever rising, not only the retrieval performance in terms of what the user is presented with as a result but also the retrieval speed proves to be crucial. Images are arguably the most prominent type of media that is to be searched and retrieved, and similar to other media such as videos or sound, images are usually stored in compressed form in order to save storage and bandwidth resources. One way of addressing the speed problem is to perform image retrieval directly in the compressed domain thus saving the time it takes to decode the pictures. In this paper we introduce such a compressed domain image retrieval technique based on the Colour Visual Pattern Image Coding (CVPIC) compression method. A combined colour co-occurrence/shape matrix is built and used in retrieval experiments. These experiments show that our approach is not only desirable from an efficiency point of view but that it also clearly outperforms popular image retrieval techniques such as colour histograms, colour coherence vectors and colour correlograms.