Visual quality indices and lowquality images

Abstract

Visual quality indices are frequently used instead of human evaluation for the quality assessment of impaired images (or video material). These visual quality indices are in turn evaluated on databases containing impaired images in conjunction with a score given by evaluation with human observers. The fitness of these indices are judged on the entire quality scale of the respective database. However, this leads to the incorrect assumption that these quality indices perform well over the whole range possible qualities. This is unfortunately not true, especially towards the low quality range of images these quality indices often show little actual correlation to human judgement. In this paper a number of visual quality indices will be evaluated with regard to the lower quality spectrum of impairments and it will be shown that the overall fitness of a quality index is not generally related to its performance regarding high impairment.