Scalable image quality assessment based on structural vectors

Image quality assessment is useful in many visual processing systems and a great deal of research effort has been put in during the recent years to develop objective image quality metrics that correlate well with the perceived quality measurement. Assessing visual quality of images is not easy since the Human Visual System (HVS) is complicated and difficult to be modelled. It is well known that the HVS is sensitive to spatial frequencies and structure in images, so accounting for structure degradation in images is essential for effective picture quality prediction. In this paper, we propose the use of singular vectors out of Singular Value Decomposition as effective structuring elements in images and use them to quantify the loss in structural information in images. The scalability of the proposed metric has been also explored since singular vectors are ordered according to their visual significance. The proposed metric has been validated convincingly on three independent databases (a total of 1196 images of different distortion types and extents), and found to outperform the relevant existing image quality metrics in literature with all circumstances.