A versatile blind JPEG image quality assessment method

It is well known that the images suffer from blocking artifacts when compressed by JPEG at low quality factors. Since JPEG uses a standard block size of 8×8 pixels, then the locations of blocking artifacts are known and occur at the boundaries of multiples of 8th pixel positions in horizontal and vertical directions. There exist a number of state-of-the-art algorithms capable of estimating the blockiness in the image for such situations. However, when an image is arbitrarily cropped or resized, the locations of block boundaries and block size are unknown and the existing block-based algorithms either fail or are not robust. For such situations, a simple no-reference method in pixel-domain is proposed in this paper. It is based on the observation that the ratio of inter-block pixel difference to the average of the two neighbouring intra-block pixel differences is likely to be higher for images compressed at lower quality factor. Therefore, the average of this ratio computed across the image is an indicator of strength of blockiness and is used as a quality metric. Unlike existing algorithms, the proposed method can be applied to images with known as well as unknown block boundaries and is robust to block misalignment making it capable of measuring the blockiness even in arbitrarily cropped and resized images. The Spearman’s-rank-order-correlation-coefficient between the quality score and subjective rating of images is computed for seven standard image databases. It is observed that the proposed method gives consistent performance with higher correlation values than most of the competitive state-of-the-art blockiness measuring algorithms. The computational complexity analysis shows that despite the high accuracy, the proposed method has lower complexity.