Image compression for indexing based on the encoding cost map

The present paper proposes a hybrid content-based image search and retrieval technique based on the use of the DC coefficient information and the Encoding Cost Map (ECM), extracted from compressed JPEG data. The proposed scheme combines color-based indexing of the DC DCT coefficient data with shape-based indexing of the ECM. The use of ECM along with the information carried in the DC DCT coefficients was shown to result to very fast and accurate search and retrieval techniques. Additionally, an indexing oriented rate-distortion technique for the optimal thresholding of AC coefficients was introduced and tested. This technique selects the AC coefficients of each block that are going to be coded in order to achieve better indexing results, while producing JPEG compliant files.