Low frequency coefficient prediction for image coding

It has been shown that about 80% of the DC coefficients of a transformed image can be discarded at the encoder and then restored at the decoder using a DC coefficient restoration (DCCR) algorithm. This approach can improve the compression rate of JPEG by over 10% at low bit rates. We extend the DCCR algorithm not only to restore the DC coefficients but also to predict some of the (0, 1)th and (1, 0)th AC coefficients. The prediction thus allows these AC coefficients to be represented using less bits. Experimental results show that, almost in all cases, images coded using the proposed algorithm have higher PSNR and better visual quality than those coded using only the DCCR.