摘要
Image Quality Assessment （IQA） is a fundamental problem in image processing. It is a common principle that human vision is hierarchical： we first perceive global structural information such as contours then focus on local regional details if necessary. Following this principle, we propose a novel framework for IQA by quantifying the degenerations of structural information and region content separately, and mapping both to obtain the objective score. The structural information can be obtained as contours by contour detec- tion techniques. Experiments are conducted to demonstrate its performance in comparison with multiple state-of-the-art methods on two large scale datasets.
Image Quality Assessment （IQA） is a fundamental problem in image processing. It is a common principle that human vision is hierarchical： we first perceive global structural information such as contours then focus on local regional details if necessary. Following this principle, we propose a novel framework for IQA by quantifying the degenerations of structural information and region content separately, and mapping both to obtain the objective score. The structural information can be obtained as contours by contour detec- tion techniques. Experiments are conducted to demonstrate its performance in comparison with multiple state-of-the-art methods on two large scale datasets.
