一种基于知觉模拟的全参考图像质量评价方法  

A Full-reference Image Quality Assessment Method Based on Perception Simulation

作　　者：李杭原 史再峰 曹清洁[2] 李金卓 黄泳嘉 Li Hangyuan;Shi Zaifeng;Cao Qingjie;Li Jinzhuo;Huang Yongjia(School of Microelectronics,Tianjin University,Tianjin 300072,China;School of Mathematical Sciences,Tianjin Normal University,Tianjin 300387,China;Tianjin Key Laboratory of Imaging and Sensing Microelectronic Technology,Tianjin 300072,China)

机构地区：[1]天津大学微电子学院,天津300072 [2]天津师范大学数学科学学院,天津300387 [3]天津市成像与感知微电子技术重点实验室,天津300072

出　　处：《南开大学学报（自然科学版）》2021年第6期12-16,22,共6页Acta Scientiarum Naturalium Universitatis Nankaiensis

基　　金：国家自然科学基金(61674115);天津市自然科学基金(17JCYBJC15900)。

摘　　要：为提高客观图像质量评价与主观图像质量评价的一致性,提出一种模拟人类感知图像质量过程中,由生理因素主导的前期过程和由心理因素主导的后期过程的IQA方法.人类感知图像信息的过程依赖于人脑的内部生成机制,它主动预测图像中易被感知的简单结构且忽略难以被感知的复杂结构.此方法利用贝叶斯预测模型将输入图像分为易感知区域和难感知区域,对具有各异感知特征的两区域通过不同特征进行分别评估.最后采用关于视觉显着性和亮度掩蔽的整体综合策略,对两组评估结果统一进行计算.基于整体感知的综合策略和基于局部感知的保真度测量在此方法中互为补充.TID2008等4个通用评测数据库的实验结果,指出了该方法具有良好的预测精度和稳定性.In order to improve the consistency between objective image quality assessment and subjective image quality assessment(IQA),an IQA metric was proposed.The metric simulated the early process dominated by physiological factors and the later process dominated by psychological factors in the process of perceiving image quality.The process of perceiving image information depended on the internal generative mechanism of brain,it actively predicted simple structure that is easy to be perceived and neglects the complex structure that is uneasy to be perceived in the image.The input image was divided into the region that is easy to be perceived and the region that is difficult to be perceived by using Bayesian prediction model.The two regions with different perceptual characteristics were evaluated by different features.The two sets of assessment results were finally combined by a comprehensive strategy on visual salience and luminance masking.The comprehensive strategy based on global perception and the local fidelity measurement based on local perception were complementary to each other in the proposed metric.According to the experimental results of four universal databases such as TID2008,the proposed metric has been proved to have good prediction accuracy and stability.

关 键 词：图像质量评价 知觉模拟 内部生成机制 视觉掩蔽 感知掩蔽 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]