基于Gabor滤波的并行网络无参考图像质量评价  

No-reference image quality assessment of parallel network based on Gabor filtering

作　　者：李国涛 杨忠[1] 张驰 朱傥 许昌亮 LI Guotao;YANG Zhong;ZHANG Chi;ZHU Tang;XU Changliang(College of Automation Engineering,Nanjing University of Aeronautics and Astronautics,Nanjing 211106,China)

机构地区：[1]南京航空航天大学自动化学院,江苏南京211106

出　　处：《应用科技》2023年第5期25-30,36,共7页Applied Science and Technology

摘　　要：为解决在基于深度学习的无参考图像质量评价(no-reference image quality assessment,NR-IQA)领域内图片失真可能导致的图片边缘模糊问题,提出一种基于Gabor滤波的并行网络无参考图像质量评价算法。考虑到图片失真的非均匀性,利用可提取空间局部频域特征的Gabor滤波器来获取边缘图片;将灰度图片和边缘图片共同输入并行网络,融合并行网络所提取到的图像质量特征,并通过全连接层所组成的回归网络将其映射到质量分数。该算法采用128×128的图片块,既能达到增加数据集的作用,又不过多损失图片的整体信息。本文在LIVE和TID2013共2个图像数据集上进行了实验,并与其他性能良好的无参考图像质量评价算法进行了对比。实验结果表明,所提出的算法与人的视觉系统(human visual system,HVS)具有较高的一致性。For solving the defect of blur of picture edge caused by image distortion in the field of no-reference image quality assessment(NR-IQA)based on deep learning,this paper presents a parallel network no-reference image quality assessment algorithm based on Gabor filtering.Considering the non-uniformity of picture distortion,the edge picture is obtained by using Gabor filter that can extract spatial local frequency domain features,then the gray picture and edge picture are input into the parallel network together,the image quality features extracted by the parallel network are fused and mapped to the quality score via the regression network constituted by all connection layers.The algorithm uses 128×128 picture blocks,which can not only increase the data set,but also prevent too much loss of the overall information of a picture.The experiments have been carried out on LIVE and TID2013 image data sets.The results are compared with other NR-IQA algorithms having good performance.Experimental results show that the proposed algorithm is highly consistent with human visual system.

关 键 词：无参考图像质量评价 图片失真 频域特征 GABOR滤波 并行网络 回归网络 深度学习 人的视觉系统 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]