基于CSR-MCA的图像融合方法  

Image fusion method based on convolutional sparse representation and morphological component analysis

作　　者：李鑫翔 张龙波[1] 王雷[1] 周晓宇 LI Xinxiang;ZHANG Longbo;WANG Lei;ZHOU Xiaoyu(College of Computer Science and Technology,Shandong University of Technology,Zibo,Shandong 255000,China)

机构地区：[1]山东理工大学计算机科学与技术学院

出　　处：《智能计算机与应用》2019年第6期24-28,31,共6页Intelligent Computer and Applications

基　　金：国家自然科学基金(61502282);山东省自然科学基金(ZR2015FQ005);山东省高等学校科技计划项目(J18KA362)

摘　　要：为了解决图像融合过程中图像信息重影失真的缺点,提出了基于卷积稀疏表示(convolutional sparse representation,CSR)和形态成分分析(morphological component analysis,M CA)的图像融合方法。利用卷积稀疏表示的优越性对形态成分分析模型进行改进,形成CSR-MCA的新型模型,可以同时实现源图像的多组件和全局稀疏表示。使用预学习的CSR-MCA模型得到源图像的平滑和细节成分的稀疏表示,然后使用不同的融合规则对每个图像分量进行融合,利用相应的字典对融合后的分量进行叠加重构获得最终的融合图像。实验结果表明,相比传统图像融合方法,本文提出的方法在主观上能很好地保留图像信息,并减少重影和失真的产生;在客观评价上,其在标准差、互信息、熵、平均梯度、空间频率等指标上表现更为优越。In order to solve the shortcomings of image information shadow distortion in the process of image fusion,an image fusion method based on convolution sparse representation( convolutional sparse representation,CSR) and morphological component analysis( morphological component analysis,MCA) is proposed. Based on the advantages of convolution sparse representation,the morphological component analysis model is improved to form a new model of CSR-MCA,which can realize the multi-component and global sparse representation of the source image at the same time. The pre-learning CSR-MCA model is used to obtain the smoothing of the source image and the sparse representation of the detail components,and then different fusion rules are used to fuse each image component,and the corresponding dictionary is used to superimpose and reconstruct the fusion component to obtain the final fusion image. The experimental results show that compared with the traditional image fusion method,the proposed method can keep the image information subjectively and reduce the generation of double shadow and distortion,and it is superior in the standard deviation,mutual information,entropy,average gradient,spatial frequency.

关 键 词：图像融合 稀疏表示 卷积稀疏表示 形态成分分析 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]