Comparison of Three Objective Image Quality Measures' Performances

The amount of distortion that is introduced by lossy image compression techniques, is evaluated mostly by objective image quality measures in the literature, because performing subjective tests is a difficult and time consuming process. In this study, three different objective image quality measures' performances are evaluated. These quality measures are mean square error (MSE), derivative signal to noise ratio (DSNR) and multiresolution distance measure (MDM). Different measures respond with differing sensitivities to artifacts and distortions. Mean square error responds more to additive noise. Derivative SNR is more sensitive to distortions in edge regions. MDM measures distortions at different resolutions and assigns larger weights to low resolutions. The performances of these measures are evaluated by considering their matches with subjective evaluations. Analysis of variance (ANOVA) technique is used for statistical analysis of these objective quality measures.