基于膨胀的梯度结构相似度图像质量评价方法  

Gradient Structural Similarity Image Assessment Index Based on Dilation

作　　者：桑庆兵[1] 梁狄林 吴小俊[1] 李朝锋[1] 

机构地区：[1]江南大学物联网工程学院计算机系,无锡214122

出　　处：《计算机科学》2014年第6期287-290,298,共5页Computer Science

基　　金：国家自然科学基金(61170120;60973094);江苏省自然科学基金(BK2011147);青年基金(61103128)资助

摘　　要：传统的梯度结构相似度算法(GSSIM)简单地将各子块GSSIM的平均值作为整幅图像的质量评估值,忽略了人眼对图像不同失真区域的视觉灵敏度不同的特点。针对此问题,提出了一种基于膨胀和图像块分类的加权梯度结构相似度图像质量评价方法(WGSSIM)。该方法首先将失真图像划分为两个区域:边缘膨胀区域和平滑区域;然后将失真图像划分成8×8的图像块,根据失真区域将图像块区分为边缘膨胀块与平滑块两类;最后对不同类型图像块之间的GSSIM值赋予不同的权值,计算得到整幅图像的WGSSIM。实验表明,该方法在3个数据库上的评价结果稳定、合理,更加符合人眼视觉系统特性,评价结果与主观评价有很好的一致性。The traditional gradient structure similarity algorithm （GSSIM） simply takes the average of each sub-block GSSIM index as quality evaluation of the whole image.The human visual sensitivity is different when observing the different areas,which is ignored by GSSIM.So an approach of weighted gradient structural similarity based on dilation and image block classification was proposed for image quality assessment.In our new method,firstly the distorted image is divided into two regions：edge dilation region and smooth region.Then the distorted image is divided into 8 × 8 image blocks,which are classified into edge dilation blocks and smooth ones according to the distorted region.The GSSIM index is given different weight values according to different type blocks.The whole image quality is calculated by Weighted GSSIM index.Experimental results on three simulated databases show that the proposed metric is more reasonable and stable than other methods.It obtains high correlations with subjective quality evaluations and low calculation,and is more consistent with human visual system.

关 键 词：图像质量评价 全参考 梯度结构相似度 人眼视觉系统(HVS) 膨胀 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]