摘要
多曝光融合(Multi-exposure Fusion,MEF)可将同一场景拍摄的几幅不同曝光图像融合成细节更丰富的图像。针对不同融合算法得到的MEF图像可能存在纹理丢失、结构混乱、亮度及颜色失真等现象,提出了一种基于非下采样剪切波变换(Non-subsampled Shearlet Transform,NSST)的多曝光融合图像质量评价方法。所提出算法先利用非下采样剪切波变换能较好地反映图像的纹理信息,获得较稀疏的图像表示,在NSST域对多曝光图像源序列和融合图像进行纹理相似度度量;其次,考虑融合图像可能会出现亮度和色度失真,提取其亮度和色度特征用于质量评价;再利用梯度特征来描述人类视觉较为敏感的边缘信息,并进行结构相似度度量。最后,采用随机森林的方法对以上4类特征进行训练,建立质量评价模型。实验结果表明,所提出方法的Pearson线性相关系数约为0. 902,Spearman秩相关系数约为0. 874,优于现有的MEF图像质量评价方法,与人眼视觉感知的一致性更好。
Multi-exposure fusion( MEF) can fuse several different exposure images taken in the same scene into more detailed images. For the MEF images obtained by different fusion algorithms may have some problems such as texture loss,structure disorder,luminance and color distortion. A new multi-exposure fusion image quality assessment method is proposed based on non-subsampled shearlet transform( NSST). Firstly,NSST is used to better reflect the texture information of the image,a sparse image representation is obtained and the texture similarity is measured in the NSST domain. Secondly,considering the fusion image may have brightness and chromaticity distortion,its brightness and color characteristics are extracted for quality evaluation. Then,the gradient feature is used to describe the sensitive edge information of human vision,and the similarity of the structure is measured. Finally,the above four types of characteristics are trained using the random forest method to establish a quality evaluation model. The experimental results show that the Pearson linear correlation coefficient is about 0. 902 and the Spearman rank correlation coefficient is about 0. 874,which is superior and more consistent with the human visual perception,in comparison with some stateof-art methods.
作者
王丹
郁梅
白永强
马华林
姜浩
WANG Dan;YU Mei;BAI Yongqiang;MA Hualin;JIANG Hao(College of Information Science and Engineering,Ningbo University,Ningbo 315211,China;National Key Lab of Software New Technology,Nanjing University,Nanjing 210093,China)
出处
《激光杂志》
北大核心
2019年第1期92-97,共6页
Laser Journal
基金
国家自然科学基金(No.61671258)
浙江省自然科学基金(No.LY15F010005)
关键词
多曝光图像融合
质量评价
纹理相似性
结构相似性
非下采样剪切波变换
Multi-exposure image fusion
quality assessment
texture similarity
structure similarity
non-sub-sampled shearlet transform
作者简介
通信作者:郁梅(1968-),女(汉族),江苏无锡人,教授,主要研究工作是多媒体信号处理与通信。E-mail:yumei2@126.com.
