一种基于特征相似性的印刷图像质量评价方法  

A Method for Evaluating the Quality of Printed Image Based on Feature Similarity

作　　者：曾台英[1] 邵雪[1] 汪祖辉[1] 

机构地区：[1]上海理工大学,上海200093

出　　处：《包装工程》2016年第11期153-157,共5页Packaging Engineering

基　　金：新闻出版总署数字印刷工程研究中心数字传播重点实验室基金(10-00-309-000)

摘　　要：目的提出一种基于特征相似性的印刷图像质量评价方法,解决客观评价印刷图像质量的方法忽略了人眼视觉的底层特征,不能与主观评价达到很好一致性的问题。方法采用能够提取视觉底层特征的特征相似性FSIM算法,应用在印刷图像质量的客观评价中,结合主观实验,对印刷图像质量加以整体评价。结果 FSIM算法的质量评分范围在0.9~1之间,在与主观评价的相关性分析上,皮尔逊相关系数（CC）为0.9490,斯皮尔曼秩相关系数（SROCC）为0.9940,平均绝对值误差（MAE）为0.0049,相较于MSE,PSNR,SSIM和MS-SSIM算法表现出更好的预测稳定性、单调性和一致性。结论实验结果表明,FSIM算法与人眼主观质量评价有着更好的一致性,更接近人类视觉系统,能够有效地进行印刷图像质量评价。The objective evaluation method of printing image quality has not achieved good agreement with the subjective evaluation, and the bottom layer of the human vision is ignored. The characteristic visual similarity algorithm FSIM, which could extract the low-level features of the human vision, was applied in objective evaluation of the quality of the printed image, and used to evaluate the overall quality of the printed image in combination with subjective experiments. The mass fraction of FSIM algorithm was in the range of 0.9~1, in the correlation analysis with subjective evaluation, the CC value was 0.9490, the SROCC value was 0.9940, and the MAE value was 0.0049. Compared to MSE, PSNR, SSIM and MS-SSIM algorithms, the performance of this algorithm was better in stability, monotonic function and consistency. Experimental results showed that FSIM algorithm had a better consistency with subjective quality evaluation of the human eye, which was closer to the human visual system, and could effectively evaluate the quality of the printed image.

关 键 词：印刷图像质量 人类视觉系统 特征相似性 主观评价 

分 类 号：TS801.3[轻工技术与工程] TS807