Spatial-spectral feature analysis in JPEG and JPEG-2000

Images are often compressed using algorithms such as JPEG and JPEG-2000. Retrieving these kinds of compressed images typically requires processing them into the uncompressed spatial form in which features are extracted for further analysis. This approach incurs many pre-processing operations, especially for large image archives. The paper studies the common features in different compressed domains so that indexing can be done directly from their respective domains. By employing a subband filter model, filters in the block-based discrete cosine transform (BDCT) and wavelet transform (WT) can be directly compared. We have found that both filters share similar characteristics. In fact, the lowpass filters in the two transforms are the same for the Haar wavelet. Therefore, due to the filter similarity, similar features can be extracted from the BDCT and WT subband outputs. We have also studied the effect of compression on their similarity. Despite the high compression, high similarity can still be found. Both our theoretical and experimental studies confirm that common features can be extracted directly from the compressed domains, irrespective of the value of the compression ratio and the use of BDCT or WT.