Subjective image quality assessment: A method based on signal detection theory

Abstract

In the field of image quality assessment, to develop computerized/objective methods whose evaluations are in close agreement with human judgments becomes a main task. However, accurate evaluation of human's subjective judgments is still a problem. Tradition methods based on mean opinion score (MOS) were not accurate enough, especially for images of minor changes or distortions. The present study tried to apply signal detection theory (SDT) in the field of image quality assessment, since SDT is particularly useful in measuring the way we make decisions under conditions of uncertainty. The results of three psychophysics experiments, in which images of different watermarking strengths were used as stimuli, showed that the SDT-based method was especially useful to detect the small loss of fidelity of images. This conclusion was supported by the higher correlation between the sensitivity score, P(A), with several computerized/objective QA indexes, such as PSNR, VIF and SSIM. Detecting subtle changes of images might involve some unknown implicit mechanisms for participants did not perform well enough in full-reference framework which allowing direct comparisons of the changed image to the original one.