Wavelet packet based approach for image retrieval in compressed domains

Images are often compressed to reduce storage. In order to avoid full decoding in image retrieval, features extracted directly from transformed data have been investigated. The popularity of JPEG and JPEG2000 motivates us to investigate a fast conversion scheme from block-based discrete cosine transform (BDCT) to wavelet domain so that similar features can be extracted from the two different domains. In the proposed scheme, BDCT is first approximated to wavelet packet (WP) transform and then the synthesis WP filter bank is applied to construct wavelet subbands. Studies found that the proposed conversion scheme has a low computational complexity as compared to the baseline approach. Two algorithms are then developed for extracting features in the wavelet subbands for retrieval. Our experimental results show that the proposed algorithms provide a good retrieval performance while the complexity is kept to be low.