摘要
文章提出一种基于图像自相似性、加权梯度以及对于图像的L1/2梯度先验下的遥感图像融合框架,利用图像在不同尺度间的自相似性特征,寻找图像的相似块,通过相似块的高频细节来丰富多光谱图像的细节信息,使得最终的图像能够保持较好的光谱信息,通过加权梯度向融合图像中注入适量细节信息,避免由于注入的比例问题导致融合图像空间信息的差异,利用对图像梯度的L1/2梯度约束来约束最终融合图像的梯度分布;同时在每一层利用目标融合函数对多光谱和全色图像进行融合,通过尺度的迭代,使得最终的融合图像不仅能够保持自相似性图像中的光谱信息,还能够保证最终融合图像与全色图像间梯度的一致性和各通道差异性。实验结果表明,该文算法在主观视觉和客观评价标准上均优于其他算法。
This paper presents a fusion framework based on image self-similarity,weighted gradient and L1/2 gradient prior for remote sensing image fusion.Using the self-similarity feature of images at different scales,the similar blocks of images are searched,and then the detail information of multi-spectral images is enriched by the high-frequency details of similar blocks so that the final image can maintain better spectral information.In order to avoid the difference of fused image spatial information caused by the injected ratio,the weighted gradient constraint is used,and then the gradient distribution of the final fusion image is constrained by the L1/2 gradient constraint on image gradient.At the same time,the multi-spectral and panchromatic images are fused by using the target fusion function in each layer.Through the scale iteration,the final fused image can not only maintain the spectral information,but also ensure the consistency of the gradient between the final fusion image and the panchromatic image and the difference of each channel.Experimental results show that the present algorithm is superior to other algorithms in subjective vision and objective evaluation criteria.
作者
方帅
余楚平
FANG Shuai;YU Chuping(School of Computer and Information, Hefei University of Technology, Hefei 230601, China)
出处
《合肥工业大学学报（自然科学版）》
CAS
北大核心
2020年第4期468-473,506,共7页
Journal of Hefei University of Technology：Natural Science
基金
安徽省自然科学基金资助项目(1508085SMF222)
中央高校基本科研业务费专项资金资助项目(JD2017JGPY0011)。
关键词
图像自相似性
加权梯度
L1/2梯度约束
分层融合
遥感图像融合
image self-similarity
weighted gradient
L1/2 gradient constraint
multi-layer fusion
remote sensing image fusion
