A non-reference perceptual quality metric based on visual attention models for videos

In this paper we study image distortions and impairments that affect the perceived quality of blackboard lectures images. We also propose a novel reference free image quality evaluation metric that correlates well with the perceived image quality. The perceived quality of images of blackboard lecture contents is mostly affected by the presence of noise, blur and compression artifacts. Therefore, the importance of these impairments are estimated and used in the proposed quality metric. In this context there is no reference, distortion free, image; thus we propose to evaluate the image perceived quality based on the features extracted from its content. The proposed objective metric estimates the blockliness and blur artifacts in the salient regions of the lecture images. The use of a visual saliency models allows the metric to focus only on the distortions in perceptually important regions of the images; hence mimicking the human visual system in its perception of image quality. The experimental results show a very good correlation between the objective quality scores obtained by our metric and the mean opinion scores obtained via psychophysical experiments. The obtained objective scores are also compared to those of the PSNR.