Abstract

To rate lossy image compression techniques in terms of transmission rate, distortion induced by the compression and reconstruction process, must be measured in a way that correlates well with human perception of the distortion. Many researchers agree that the common computational distortion metrics mean-square error (MSE) and root mean-square error (RMSE) do not correlate well with human subjective judgment.¹⁻⁸ In an effort to find a fidelity metric that corresponds more closely to human fidelity perception, some researchers have experimented with other measures such as the MSE weighted by properties of the human visual system (HVS).^2,5,6,9−14 No standard computational fidelity metric has been adopted by all researchers, and because they are computationally simple, MSE based measures remain the de facto metric. Most image fidelity testing with humans has been performed using different methods or under conditions that vary from lab to lab so that it is difficult to compare one researcher’s results with those of another. Statements in the literature such as "… an average of 0.26 bits/pixel can be achieved with very little perceivable degradation" or "Good Quality images are achieved with as low as .5 bit/pel"¹⁵ make it nearly impossible to compare image compression methods presented in different papers.

© 1989 Optical Society of America