Abstract
Recently many different color image difference metrics have been proposed, to quantify visual differences between an original image and a modified version of it; some in the context of overall image quality and some to quantify specific image distortions. However, none of the proposed metrics have been able to demonstrate good and stable performance for various image distortions, that is, the metric values do not correspond well to the actual perceived differences.
A notable image difference metric, S-CIELAB, was defined by Zhang and Wandell (1996) as a spatial extension to the well-known CIE 1976 color difference equation. It introduced a spatial pre-processing in an opponent-color space, to simulate the properties of the human visual system. After filtering, the images are transformed into CIELAB color space, where a conventional CIE 1976 color difference is calculated, pixelwise.
Another recent image difference metric is defined by the hue angle algorithm, proposed by Hong and Luo (2002). This metric, which is about to be recommended in a report by the International Commission on Illumination (CIE), does not take into account the spatial properties of the human visual system, and for this and other reasons, we show that it miscalculates the perceptual difference between an original image and a modified version of it.
Based on this we propose a new color image difference metric, as an extension to the hue angle algorithm, taking into account the spatial properties of the human visual system. We have subjected the proposed metric to extensive testing on various image databases containing different image distortion types. The results show significantly improved performance compared to the metric proposed by Hong and Luo, and also good performance compared to other state of the art image different metrics.