On the optimisation of nonlinear mapping functions towards high correlation of full-reference image quality metrics and their combinations with subjective evaluations

One of the most relevant problems related to most of the objective full-reference image and video quality assessment methods is their correlation with subjective perception of various types of distortions. Since the relation between the subjective scores, typically expressed as Mean Opinion Scores (MOS) or Differential MOS, and most of the metrics is nonlinear, various mapping functions are used in order to linearize this relation e.g. logistic function recommended by the Video Quality Experts Group (VQEG). Such compensation can also be conducted using some other function e.g. polynomial or exponential ones. Nevertheless, the results obtained for each of available datasets differ and such approach cannot be considered as universal. One of possible solutions, proposed in some earlier papers, is the use of combined metrics, which have significantly higher linear correlation with MOS or DMOS values without the necessity of nonlinear mapping. The comparative analysis of the optimisation results for some “state-of-the-art” metrics and proposed combined ones is also provided in the paper.