基于改进GoDec算法的红外与可见光图像融合  

Infrared and visible image fusion based on improved GoDec algorithm

作　　者：张梦 邓小颖[1] 曹海涛 张剑云 贺翔 朱金荣 ZHANG Meng;DENG Xiaoying;CAO Haitao;ZHANG Jianyun;HE Xiang;ZHU Jinrong(School of Information Engineering(School of Artificial Intelligence),Yangzhou University,Yangzhou 225002,China;School of Electrical and Energy Engineering,Yangzhou University,Yangzhou 225002,China)

机构地区：[1]扬州大学信息工程学院人工智能学院,江苏扬州225002 [2]扬州大学电气与能源动力工程学院,江苏扬州225002

出　　处：《激光杂志》2022年第8期135-140,共6页Laser Journal

基　　金：国家自然科学基金(No.61802336)。

摘　　要：为了提高红外与可见光融合图像的边缘强度和视觉信息保真度,并考虑到源图像中的腐败噪声成分,尽量保留其中的有效信息。提出了一种基于改进的GoDec(Go Decomposition)算法的红外与可见光图像融合方法。通过基于广义最大相关熵(GMCC)的GoDec算法对源图像分解得到低秩与稀疏图像,采用基于非下采样Contourlet变换(NSCT)的方法融合低秩图像,使用加权平均策略融合稀疏图像。实验结果表明,与其他5种融合方法相比,该方法的融合图像的平均梯度提高了10.3%到54.5%,边缘强度提高了3.1%到47.6%,空间频率提高了33.3%到110%,图像清晰度提高了28.1%到69.2%,视觉信息质量提高了8%到50%。In order to improve the edge intensity and visual information fidelity of infrared and visible fusion image,and considering the corrupt noise component in the source image,the effective information is preserved as much as possible.An infrared and visible image fusion method based on the improved GoDec(Go Decomposition)algorithm is proposed in this paper.The source images were decomposed by GoDec algorithm based on generalized maximum correlation entropy(GMCC)to obtain low-rank and sparse images.The low-rank images were fused by non-subsampled Contourlet transform(NSCT),and the sparse images were fused by weighted average strategy.Experimental results show that compared with the other five fusion methods,the average gradient,edge intensity,spatial frequency,image sharpness and visual information quality are improved by 10.3%to 54.5%,3.1%to 47.6%,33.3%to 110%,28.1%to 69.2%and 8%to 50%respectively.

关 键 词：红外与可见光图像融合 广义最大相关熵 改进的Go decomposition 非下采样CONTOURLET变换 

分 类 号：TN209[电子电信—物理电子学]