基于人眼视觉的结构相似度图像质量评价方法

Method of image quality assessment based on human visual system and structural similarity

分析了数字图像中亮度、纹理细节、空间位置等因素对人眼视觉特性的影响,建立了数学模型,将人眼视觉特性与图像的结构相似度结合起来,提出一种符合人眼视觉特性的图像质量评价新方法.该方法将图像划分成大小相等的分块,计算出各分块的亮度影响因子、纹理细节影响因子和空间位置影响因子,经过归一化处理得到每个分块的权值,用加权平均的结构相似度作为图像质量的评价指标.实验证明该方法能够区别图像中不同区域的图像特征,符合人眼视觉特性,与主观评价结果一致.

Through analyzing the influence of certain digital image factors （brightness, texture details, dimentional position, etc. ） on human visual characteristic, a mathematic model which combines human visual characteristic and structural similarity of images was built, a new image-quality assessment method in accord- ance with human visual characteristic was put forward. By unoverlapped partitioning of images using equalsized sliding windows, this method can calculate the following factors with influence on the partitions： influencing brightness, influencing texture details, and influencing dimentional position. The weighted value of each partition is generated after normalization, and the weighted structural similarity is used as the assessing index of image quality. Experiment shows that this method is in accordance with human visual characteristic, and consistent with the result of subjective assessment.