摘要
针对图像中的高斯模糊失真,提出一种融合反向传播的无参考模糊图像质量评价方法。利用分水岭算法标记出的连通域计算密度差值;利用Tamura纹理特征模型和拉普拉斯算子分别度量图像的粗糙度和清晰度;将提取的密度差、粗糙度和清晰度输入反向传播(Back-Propagation,BP)神经网络进行训练,实现对高斯模糊失真图像的质量预测。实验证明,该方法在质量评价以及一致性方面均优于对比方法。此外,该方法解决了实际应用中因缺乏参考图像而不能进行质量评价的问题。
Aimed at the Gaussian blur distortion in the distorted images,a no-reference quality assessment of blur images based on back-propagation is proposed.The connected component was marked by the watershed algorithm and the density difference of them was calculated.The roughness and the sharpness of the image were extracted by the Tamura texture feature model and the Laplace operator,respectively.The density difference,the roughness and the sharpness were input to back-propagation neural network for training,so as to accurately predict the quality score of Gaussian blur image.Experimental results show that,this algorithm outperforms the comparison methods in on image quality evaluation and consistency.In addition,it solves the problem that quality evaluation cannot be carried out due to the lack of reference images in practical applications.
作者
赵月
王来花
王伟胜
乔丽娟
阮泉
Zhao Yue;Wang Laihua;Wang Weisheng;Qiao Lijuan;Ruan Quan(College of Software,Qufu Normal University,Qufu 273165,Shandong,China)
出处
《计算机应用与软件》
北大核心
2022年第9期248-254,306,共8页
Computer Applications and Software
基金
国家自然科学基金青年基金项目(61601261)
山东省自然科学基金项目(ZR2016FB20)。
关键词
BP神经网络
高斯模糊
图像质量评价
无参考
连通域
粗糙度
清晰度
Back-Propagation(BP)neural network
Gaussian blur
Image quality assessment(IQA)
No-reference
Connected component
Coarseness
Sharpness