No-reference blur image quality assessment based on gradient similarity

With the popularity of the consumer electronic products, such as cell phone and other low-cost digital cameras, a large number of digital images are generated to promote interest in the study of the no-reference objective image quality assessment algorithm. In this paper, based on the extracted edge dilation block from blur edge dilation image, a novel no-reference image quality assessment scheme using gradient structural similarity (NRGSIM) is proposed for quality evaluation of blurred images. In this method, firstly the re-blurred image is produced by blurring the original blurred image with a low pass filter. Then the edge dilation image is divided into 8 × 8 blocks. The sub-blocks are classified into edge dilation block and smooth block. The gradient structural similarity index is given by different weighs according to different types of blocks. Finally, the blur estimation of the whole image is produced. Experimental results on four open blur image databases show that the proposed metric is more reasonable and stable than other methods. It obtains high consistence with subjective quality evaluations and has easy calculation. It is more consistent with human visual system. So the proposed metric is appropriate for no-reference blurred image quality assessment. The index of SROCC on LIVE2 database is 0.9641.