Based on characteristics of HVS (human visual system), a new objective image quality assessment is proposed. In this method, the information of luminance, frequency and edge is used to predict the quality of compressed images. Multi-linear regression analysis is used to integrate these information. Experimental results show that this new image quality assessment closely approximates human subjective tests such as MOS (mean opinion score) with high Pearson and Spearman correlation coefficients of 0.970 and 0.976, which are of significant improvement over some typical objective image quality estimations such as PSNR (peak signal-to-noise ratio).