Abstract
An image index, extracting DCT domain color and texture features in indexing process, is proposed for JPEG and MPEG fast image retrieval. Feature extraction from 1,000 JPEG images requires 2.86 seconds, which is 116.17 times faster than previously reported by Serata (2006) because of efficient compressed domain feature extraction. Retrieval performances of the proposed and Serata's indexes are evaluated on Corel database, and the results suggest that the maximum precision and recall of the proposed index are improved by 22.7% and 11.1% with reduction of indexing times. Due to reduced computational times, the index is applicable to databases which include 100 times larger data sizes than current one.
