Region-based image retrieval in JPEG compressed domain

Abstract

In this paper, we propose three algorithms of region-based image retrieval in the compressed domain. The unique feature of our algorithms is that the indexing of the images is done directly in their bit streams rather than DCT coefficients or pixels. The fact that decompression is completely eliminated clearly marks the advantage of our algorithms. Another notable feature of our system is that it eliminates the problem that users might have in selecting individual regions during the query process. The first implementation involves generating multiple keys for an image whereby each key is derived from the individual regions. In the second algorithm, the index key generated is derived from the most influential region, whereas for the third algorithm, the key is constructed based on weights derived from each region. Results are presented to demonstrate the effectiveness of our approaches.