A Human Perception Based Performance Evaluation of Image Quality Metrics

Though numerous image quality measures have been proposed, the search for a reliable IQM is still vigorously pursued by different research groups around the world. There is a need to compare the already proposed IQMs with respect to their adherence to human image quality perception. A model that can accurately simulate the human perception of image quality is a challenging task due to limited human knowledge in the related domains of psychology, vision, biology etc. The psycho-visual experiments remain the most accurate way to model human perception of visual quality. In this paper, different state of the art full-reference objective image quality metrics (IQMs) are evaluated against human subjective judgments on standard LIVE image quality database. The difference mean opinion scores (DMOS) were calculated from 17400 human judgments on 348 images distorted with white noise, Gaussian blur and Rayleigh fast-fading distortions. Subsequently, 13 leading IQMs like SSIM, VIF, FSIM, etc. were compared with DMOS on the basis of Pearson correlation coefficient. It is observed that though there is not a single winner, VIF and IFC seem to have a higher performance compared to other quality metrics.