No-reference perceptual blur metric for stereoscopic images

In this paper, we propose a no-reference perceptual blur metric for 3D stereoscopic images. The proposed approach relies on computing perceptual local blurriness map for each image of the stereo pair. To take into account the disparity/depth masking effect, we modulate the obtained perceptual score at each position of the blurriness maps according to its location in the scene. Under the assumption that, in case of asymmetric stereoscopic image quality, 3D perception mechanisms place more emphasis on the view providing the most important and contrasted information, the two derived local blurriness maps are combined using weighting factors based on local information content. Thanks to the inclusion of those psychophysical findings, the proposed metric handles efficiently symmetric as well as asymmetric distortions. Experimental results show that the proposed metric correlates better with human perception than state-of-the-art metrics.