No Reference Quality Assessment of Blurred Images

The overall quality of an image gets affected due to existence of out-of-focus-blur which is very common during photo capturing. In order to know the quality of the blurred image automatically and accurately, a no-reference method is proposed in this paper. The proposed method is based on the fact that if the image that is to be assessed is blurred intentionally, the effect on its structure will vary for different blurriness levels. The structure of image having no traces of blur is altered much more on deliberate blurring than the image which is highly blurred. To measure this structural change, SSIM is employed which gives the structural similarity of the image before and after willful blurring. The structural similarity value so obtained is trained and tested against the ground truth of the image by a machine learning regression algorithm. The proposed method is highly accurate giving Pearson linear correlation coefficient (PLCC) and Spearman rank correlation coefficient (SRCC) value of 0.99 and 0.97 respectively.