基于非下采样轮廓波变换的模糊图像质量评价  

Blurred image quality assessment based on non-subsampled contourlet transform

作　　者：文继李 丁立新[1] 万润泽[1] WEN Ji-li;DING Li-xin;WAN Run-ze(School of Computer Science,Wuhan University,Wuhan 430072,China)

机构地区：[1]武汉大学计算机学院,湖北武汉430072

出　　处：《计算机工程与设计》2018年第4期1078-1081,1201,共5页Computer Engineering and Design

基　　金：湖北省自然科学基金面上基金项目(2015CFB405);湖北省教育厅科学技术研究基金项目(Q20153003);湖北省高等学校省级教学研究基金项目(2016419)

摘　　要：为判断模糊图像的失真程度,针对模糊图像的质量评价提出一种方法。通过非下采样轮廓波变换提取图像特征形成特征向量,将得到的特征向量作为输入,图像的DMOS值作为输出,利用支持向量回归建立图像质量评价模型,结合序列后向选择算法预测图像质量得分。实验结果表明,在LIVE、VCL@FER、CSIQ和TID2013数据集上,该方法和人类视觉系统具有较好的一致性。To judge the distortion degree of blur image,a method was proposed for the quality assessment of blurred image.Image features were extracted by NSCT to form feature vector,and the feature vector was taken as input,while the DMOS value of the image was used as output.The image quality assessment model was established using support vector regression(SVR),and the sequence backward selection(SBS)algorithm was combined to predict the image quality score.Experimental results show that the proposed method and human visual system(HVS)were verified to be in consistency on LIVE,VCL@FER,CSIQ and TID2013 datasets.

关 键 词：模糊 非下采样轮廓波变换 支持向量回归 序列后向选择 图像质量评价 人类视觉系统 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]