We extend a new visual-error criterion for achromatic images to color images. After the tristimulus planes are transformed to a luminance-compatible color space, the achromatic-error criterion is used for the luminance plane and the squared-error criterion is used for the two chromatic planes taken together. The optimal linear estimator for the resulting color-image-error criterion is compared against a purely squared-error criterion in the task of reconstructing color images degraded by varying amounts of additive noise. The images restored with the new error criterion appeared sharper and truer in fine detail. Except for the least and most noisy cases, in which no clear preference was found, viewers of the cathode-ray-tube display unanimously preferred these restorations over the comparatively fuzzy squared-error ones, despite a grainier appearance. These results are consistent with the previous ones found with the new visual-error criterion for achromatic images only. Through this extension to color images and the good results in restoring noisy images, we have found an error criterion that shows promise for applications in color-image processing.
© 1982 Optical Society of AmericaPDF Article