A rank-order comparison of image quality metrics

This paper presents a comparison of a series of image quality metrics. Nine full-reference metrics of varying complexity are used to sort a number of different sets of corrupted images. Within each set, images are ranked from highest to lowest quality, using each of the nine metrics. The degree of similarity between each of these nine rank-orders is then statistically assessed. Results suggest that the metrics ultimately arrange distorted images in much the same way, thereby drawing a link between the sophisticated metrics and the more elementary metrics.