Coarse and Fine Structure Combined Model For Full Reference Image Quality Assessment

The structure is proven to be an effective quality indicator for full-reference (FR) image quality assessment (IQA). Existing FR-IQA methods only extract fine structure for distortion detection. This leads to error detection when the distortion is invisible. To this end, we propose a coarse and fine structure combined FR-IQA method in this paper. In specific, we design a patch based structure detection method to detect the coarse structure in an image, and detect the fine structure using the gradient magnitude (GM). The proposed method is tested on TID2013, TID2018, CSIQ and LIVE databases. The validation results demonstrate that the proposed model have high prediction accuracy, yet have modest computational complexity.