Image retrieval based on subband energy histograms of reordered DCT coefficients

Nowadays a large number of images are compressed in the JPEG format. In order to retrieve these images, firstly we should decompress these images, then extract feature vectors in the pixel domain. Some of the literature has proposed image retrieval methods directly in the DCT domain, which can reduce computation complication. In this article, we propose a novel method for image retrieval. We firstly reorder the DCT coefficients of the JPEG images in the multiresolution wavelet transform form; this can get a set of subbands. Secondly construct subband energy histograms of these subbands, then construct image keys by using Morton order, and lastly organize the image keys in the form of a variant of the B-tree data structure for image retrieval.