Capturing structural distortions in digital images and videos

With the growing demand for image and video services, objective analysis of image and video quality has received increased interest from the content providers and network operators. This study proposes to capture structural image/video distortions through spatial/spatiotemporal orientation analysis. For image quality assessment (IQA), we significantly improve the classic SSIM algorithm with low computational overhead by taking into account the preservation of edge orientations. For video quality assessment (VQA), a unified framework for attention guided structural distortion measure is presented based on the motion-tuned spatiotemporal oriented energies and a spatiotemporal visual saliency model, in which the descriptive and efficient distributed motion representation is employed to alleviate the typical problems of the commonly used optical flow methods. The structural distortion measure is then combined with a multi-scale SSIM based spatial distortion measure to form a comprehensive video distortion metric, which demonstrates good quality prediction and high computational efficiency.