Comparison of combination rules for digital-image-coding impairments

At high compression ratios, digitally-coded images usually comprise several coding artifacts, each impairing perceptual image quality to a greater or lesser extent. In this paper, two possible rules for combining such digital-image-coding impairments are presented and tested experimentally, viz., Annatt's 'law of subjective addition' and a rule based on the Minkowski p-metric. AllnaU's rule assumes a reciprocal relation between quality and impairment and a linear addition of impairments. The rule using the Minkowski p-metric assumes a linear relation between quality and impairment and a nonlinear addition of impairments. Two experiments are presented in which these rules are tested. In the first experiment, the stimuli consisted of a set of scale-space-coded images in which the sharpness of the image was systematically varied. Both image quality and impairment were assessed by means of numerical category scaling. In the second experiment, subjects rated image quality of scale-space-coded colour images comprising one to three perceptually different impairments, namely, 'unsharpness', 'phantoms' (dark/bright patches in bright/dark regions) and 'colour desaturation'. The results of the first experiment show a linear relation between quality and impairment. The combined results of the first and second experiment indicate the Minkowski p-metric as the most suitable rule for describing the observed accumulation of digital-image-coding impairments. Exponent p was found to be about two.