Direct content access and extraction from JPEG compressed images

In this paper, we propose a novel design of content access and extraction algorithm for compressed image browsing and indexing, which is critical for all visual information systems. By analyzing the relationship between DCT coe3cients of one block of 8 × 8 pixels and its four sub-blocks of 4 × 4 pixels, the proposed algorithm extract an approximated image with smaller size for indexing and content browsing without incurring full decompression. While the computing cost is signi9cantly lower than full decompression, the approximated image also reserves the content features, which are su3cient for indexing and browsing as evidenced by our extensive experiments. ? 2002 Pattern Recognition Society. Published by Elsevier Science Ltd. All rights reserved.