Masking effect of non-predictable energy of image regions
Tutkimustuotos › › vertaisarvioitu
|Julkaisu||Telecommunications and Radio Engineering|
|DOI - pysyväislinkit|
|Tila||Julkaistu - 2017|
This work relates to the area of full-reference image visual quality assessment. The paper deals with studying the impact of image fragment non-predictivity on image perception by humans, in particular, on ability of such fragments to mask distortions. A hypothesis that the fragment masking effect depends on how difficult is it for a human to predict image fragment content based on neighbor fragments is formulated. It is supposed that a larger prediction error is correlated with a larger masking effect. Experiments hold to practically confirm this hypothesis with forming test images and determining mean opinion scores, MOS, for them are described. It is shown that taking into account the considered peculiarities of human vision it becomes possible to introduce in images distortions with by five times larger power (variance) compared to the case of uniform distribution of distortions. A modification of the conventional metric PSNR that takes into account these peculiarities is proposed.