Influence of Degradation on Modified Full Reference Image Quality Metrics

This paper presents the influence of six different types of image degradation on the modified objective full-reference PSNR quality metrics. The types of degradation used are different levels of Contrast, Median filter, JPEG compression, Gaussian blur, Motion blur as well as white Gaussian noise with zero standard deviation and variable mean. The modified PSNR quality metric uses MAE, Euclidean distance and Canberra distance as error estimators instead of MSE. The results show that for all types of degradation the quality metric with the lowest deviation is PSNR modified with Euclidean distance. The Gaussian noise with mean shift is the most appropriate degradation to be estimated by all modifications, whereas median filter and contrast produce the worst results. The results by PSNR with Euclidean distance can be used with standard uncertainty when the quality is degraded by Gaussian blur, Motion blur, Median and positive mean due to normality of the data.