Abstract: For undistorted images, the wavelet coefficients of between-scale coefficients in the same orientation are correlated, while compression coding reduce the correlation. Cosine similarity is used in this work to model the correlation of between-scale subbands, and statistical regression is applied to analyze the relationship between human subjective assessment Mean Opinion Scores (MOS) and subbands cosine similarity. The accurate quality model is obtained by regression analysis. Experimental results show that the proposed no-reference method has a high correlation with the MOS measurement, and a considerably lower computational complexity and less run time.