Abstract:

Wavelet domain and structural similarity(SSIM) quality assessment method have become hotspots in the field of image processing, however, both of them have some flaws: the traditional discrete wavelet transform lacks of translational invariance and its direction selectivity is also highly limited; for severe blurred images, the results of the SSIM are not very accurate. Therefore, we propose a new algorithm for blur image quality evaluation. This algorithm uses dual tree complex wavelet transform (DTCWT) image decomposition to obtain the complex wavelet coefficients and the high frequency sub band coefficients of all the six directions. Then the structural similarity of the average gradient amplitude is measured. Finally all the mean gradientmagnitudebased structural similarity ( MGSIM) average is calculated as the final fuzzy values of the original blur images. Experimental results show that the proposed method fits the visual characteristics of human eyes better in contrast with the structural similarity method, and well matches the results of subjective evaluation methods. The assessment results are better than the current literature in terms of overall performance.

Key words: wavelet domain;dual-tree complex wavelet transform;SSIM;mean gradient magnitude based structural similarity