一种基于遮蔽效应的图像质量评价研究  

A study on image quality evaluation based on masking effect

作　　者：孔紫剑 周冬明[1] 聂仁灿[1] 王长城 KONG Zi-jian;ZHOU Dong-ming;NIE Ren-can;WANG Chang-cheng(School of Information Science&Engineering,Yunnan University,Kunming 650500,Yunnan,China)

机构地区：[1]云南大学信息学院,云南昆明650500

出　　处：《云南大学学报（自然科学版）》2021年第6期1096-1106,共11页Journal of Yunnan University(Natural Sciences Edition)

基　　金：国家自然科学基金(61966037,61463052,61365001).

摘　　要：针对梯度结构相似性指标(Gradient Structure Similarity,GSSIM)无法对近阈值失真图像做出很好的判断,导致其判断结果与人类视觉系统(Human Visual System,HVS)不完全一致的问题,为提高GSSIM的准确性及其与HVS的一致性,提出了一种基于梯度遮蔽和视觉显著性的图像质量评价指标(Visual Saliency-Gradient Structure Similarity,VS-GSSIM).首先,根据梯度信号之间存在的遮蔽效应优化梯度性相似性模型;然后,结合显著性模型提高指标和HVS评判的一致性;最后,调整图像局部区域感知质量水平的相对重要性并池化最终得分.实验结果表明所提模型在一致性和单调性的评价指标上均超过GSSIM,并且优于目前绝大多数算法.The Gradient Structure Similarity(GSSIM)cannot make a good judgment on the near-threshold distortion images,which leads to non consistent with the Human Visual System(HVS).In order to improve the accuracy of this index and its consistency with HVS,an image quality evaluation index based on gradient masking and visual saliency(VS-GSSIM)is proposed.The index first optimizes the GSSIM model according to the masking effect between the gradient signals,then combines with the saliency model to improve the consistency of the index and the HVS evaluation,and finally,it adjusts the relative importance of the perceived quality level of the local area of the image and pooling the final score.The experimental results show that the proposed model surpasses GSSIM in terms of consistency and monotonic evaluations,and is better than most current algorithms.

关 键 词：图像质量评价 人类视觉系统 梯度相似性 遮蔽效应 显著性模型 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]