Visual similarity index for image quality assessment

Low level features are widely used in computer vision for acquiring information from outside circumstance and responding to it. Considering that low level features provide a rich source of information about luminance distribution, object organization and foreground/background configuration, their difference reflects the structural change of images. Based on the fact that the human vision system always focuses on the local neighborhoods around gazing positions, similarity between corner and edge of images is estimated locally and combined into an image quality metric, namely low-level features based similarity measure (LFSIM). Extensive experiments based upon five publicly-available image databases with subjective ratings demonstrate that LFSIM performs much better than traditional peak signal noise ratio (PSNR) and structural similarity measure (SSIM), and is even competitive to the state-of-the art image quality assessment algorithms information fidelity criteria (IFC) and visual information fidelity (VIF), which are developed on the basis of natural scene statistics.