Design and evaluation of perceptual masks for wavelet image compression

Transform-based image compression performance is measured quantitatively in terms of peak signal-to-noise ratio (PSNR); subjective quality is also important to evaluating performance. An image compression system that incorporates the sensitivities of the human visual system (HVS) can enhance performance. This paper develops new HVS techniques-in the form of perceptual weighting masks derived from the luminance contrast sensitivity function-and illustrates their performance for grayscale and color images using the biorthogonal 9/7 and 22/14 wavelet transforms. Performance is measured quantitatively and qualitatively using a subjective testing procedure with 15 observers. Our HVS methods improve both PSNR and subjective quality for color images; for grayscale images, PSNR values decrease and quality increases.