No abstract

Nowadays, numerous video compression quality assessment metrics are available. Some of these metrics are "objective" and only tangentially represent how a human observer rates video quality. On the other hand, models of the human visual system have been shown to be effective at describing spatial coding. In this work we propose a new quality metric which extends the peak signal to noise ratio metric with features of the human visual system measured using modern LCD screens. We also analyse the current visibility models of the early visual system and compare the commonly used quality metrics with metrics containing data modelling human perception. We examine the Pearson's linear correlation coefficient of the various video compression quality metrics with human subjective scores on videos from the publicly available Netflix data set. Of the metrics tested, our new proposed metric is found to have the most stable high performance in predicting subjective video compression quality.

No abstract