摘要
结构相似性理论是一种关于图像质量评价的新思想.与自底向上地模拟人眼视觉系统(HVS)低阶的组成结构不同,结构相似性理论自顶向下地模拟HVS的整体功能.作为结构相似性理论的一个实现,结构相似度(SSIM)指数有着简单高效的优点,但SSIM在交叉失真类型和失真严重时的准确性不够好.本文将结构信息重新解释为图像中能量足够大的中高频成分,从新的角度将SSIM理解为一种更好的局部误差度量方式.提出一种基于结构信息提取(SIExt)的图像质量评价方法,将图像中的结构信息分离出来给予较大的权重,并用SSIM作为误差度量估计局部失真.实验结果表明,SIExt比PSNR和SSIM等方法有更好的准确性.
The philosophy of structural similarity is a new idea about image quality assessment.Different from the traditional paradigm which modeling the low level composition of Human Visual System （HVS） bottom-up, the new philosophy modes the functionality of the overall HVS from top to down.As an implementation of the new philosophy,the Structural SIMilarity （SSIM） index is simple and efficient. However, SSIM fails in measuring the badly distorted images and images with cross distortion types. In this paper, we interpret the structural information as the mid and high frequencies with enough energy, and regard the SSIM index as a better distortion measure of local areas. Then, a new image quality index based on Structural Information Extraction （SIExt） is proposed. In SIExt, structural information is separated from image and given higher weight, and SSIM is used as an error measure to estimate local distortions. Experimental results show that the proposed SIExt can assess the quality of images more accurate than PSNR and SSIM.
出处
《电子学报》
EI
CAS
CSCD
北大核心
2008年第5期856-861,共6页
Acta Electronica Sinica
基金
北京市自然科学基金(No.4072004)
北京市教委科技发展计划基金(No.KM200510005012)
