Digital image compression using directional sub-block DCT

A digital image compression technique using a directional sub-block discrete cosine transform (DCT) is presented. The proposed technique gives us a reconstruction image with lower block artifact (noise) even if a low-bit-rate coding is applied. The DCT signs and amplitudes are processed separately. DCT amplitudes of one sub-block are predicted from those of one of the former four neighbor sub-blocks and the prediction errors are coded. The one of four neighbor sub-blocks is chosen by a minimum mean square error (MMSE) technique. The indications of the chosen sub-block and DCT signs are coded by an arithmetic coding. The picture quality of the reconstructed image is always better than that obtained by an ordinary JPEG method.