Image quality assessment metrics combining structural similarity and image fidelity with visual attention

Abstract

Image quality assessment has a great importance in several image processing applications. Recently, various objective image quality metrics have been proposed in order to predict human visual perception. In this paper, novel image quality metrics, S-SSIM (saliency-based structural similarity index) and S-VIF (saliency-based visual information fidelity), are proposed based on a visual attention model extracting frequency-tuned salient region. Saliency maps are produced from the color and luminance features of the image. SSIM and VIF in pixel domain are modified by the weighting factors of the saliency maps. We validated our approach using 2 image databases as test bed: These databases contain subjective scores for each image. Our results showed that our technique is more correlated with human subjective perception.