Abstract: Most of the current image compression algorithms work on the purely-pixel-value-based mean square error （MSE） criteria.But MSE criteria can’t describe the difference between the edge and flat area in image.Thus the optimal algorithms based on MSE criteria may not achieve good results.A improved MSE criteria is proposed.Based on the characteristics of the human vision,the new approach takes into consideration the statistical characteristics of each subband and the different effects of subband coefficients on saving the edge and texture information of original image.Experimental results show that this approach can obtain satisfying results,save the edge and texture information well under low bit rate.In addition,the subjective quality of the reconstructed image is improved to some extent.