基于梯度幅度和梯度方向直方图的全参考图像质量评价算法  

Full-reference image quality assessment algorithm based on gradient magnitude and histogram of oriented gradient

作　　者：王同罕 贾惠珍 舒华忠[2] Wang Tonghan;Jia Huizhen;Shu Huazhong(Jiangxi Engineering Laboratory on Radioactive Geoscience and Big Data Technology,East China University of Technology,Nanchang 330013,China;Laboratory of Image Science and Technology,Southeast University,Nanjing 210096,China)

机构地区：[1]东华理工大学江西省放射性地学大数据技术工程实验室,南昌330013 [2]东南大学影像科学与技术实验室,南京210096

出　　处：《东南大学学报（自然科学版）》2018年第2期276-281,共6页Journal of Southeast University：Natural Science Edition

基　　金：国家自然科学基金资助项目(61762004);江西省教育厅科学技术研究资助项目(GJJ170427);东华理工大学江西省放射性地学大数据技术工程实验室开放基金资助项目(JELRGBDT201702);东华理工大学博士启动基金资助项目(DHBK2016119;DHBK2016120)

摘　　要：为了提高全参考图像质量评价算法的性能,提出了一种基于梯度幅度和梯度方向直方图的图像质量评价算法.梯度方向直方图可以描述图像中局部物体表象和形状,梯度幅度能精细地反应图像中微小细节的反差和纹理变化.分别计算参考图像和失真图像的梯度幅度和梯度方向直方图,然后计算梯度幅度相似度图和梯度方向直方图相似度图,最后通过标准方差加权的方式得到图像的预测质量分数.在LIVE,TID2008和IVC三个图像数据库上的实验结果表明,所提算法的预测结果与人的主观判断具有较好的一致性.To improve the performance of full-reference image quality assessment(IQA)algorithms,an IQA algorithm based on the gradient magnitude(GM)and the histogram of the oriented gradient(HOG)is proposed.The HOG can characterize the representation and the shape of the local object,and the GM can finely reflect the potential tiny changes in texture and detail.The gradient magnitude and the histogram of the oriented gradient of the reference image and the distorted image are computed.Then the similarity maps of the gradient magnitude and the histogram of the oriented gradient are calculated.Finally,the predictive quality score is obtained by using the standard deviation method.The experimental results on LIVE,TID2008 and IVC databases show that the proposed algorithm has a good performance in term of correlation with human judgments.

关 键 词：图像质量评价 全参考 梯度幅度 梯度方向直方图 

分 类 号：TN911.73[电子电信—通信与信息系统]