Colour difference metrics and image sharpness

Abstract

We describe experiments to assess differences in the perceived sharpness of images, in which a Gaussian filter of variable width was convolved with a set of test images. Differences between the original and filtered images were computed pixel by pixel using three colour metrics: (a) mean square error (MSE) computed in RGB values; (b) CIELAB colour difference; and (c) s-CIELAB spatial colour difference. A psychophysical experiment was performed in which observers assessed the sharpness of the filtered images on a numerical category scale. The modulation transfer function (MTF) for the display was measured and the square root integral (SQRI) was evaluated as a predictor of perceived image sharpness. The results showed good agreement between the mean observer judgements and the prediction of both the colour difference metrics and the image quality metric across the full range of widths of the Gaussian filter.