Unified feature analysis in different compressed domains

Images can be compressed using different algorithms such as JPEG and JPEG 2000. Retrieving these kinds of images typically require processing them onto 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 objective of this paper is to study features that are common in these domains so that indexing can be done directly from their compressed domain. By deriving expressions for the lowpass filters in both block-based discrete cosine transform (DCT) and wavelet transform (WT), it is found that these two filters are very similar. In fact, it is shown that they are the same for the Haar wavelet kernel. Both theoretical and experimental studies show that features that are common to DCT and WT domains can be extracted. Also, the effect of compression ratios on these two formulations has been studied. It is found that despite the high compression ratio, features that are extracted from these two compressed domains still can have a large similarity.