摘要
针对在图像融合中存在边缘细节保留不够理想的问题,提出一种基于非下采样剪切波变换(NSST)与卷积神经网络图像融合框架(IFCNN)的红外可见光图像融合算法。首先将红外和可见光图像进行NSST分解。然后为了使低频子带图像更好地突出轮廓信息,使用相似性匹配的融合规则对图像进行融合;对高频子带图像使用IFCNN提取特征层,特征层通过L2正则化、卷积运算和最大选择策略处理可以得到最大权重图,根据最大权重图来确定高频融合规则。最后使用NSST逆变换得到最终的融合图像。实验结果表明,所提算法很好地保留图像的边缘及纹理等细节信息,减少伪影和噪声,具有良好的视觉效果。
Aiming at the problem of insufficient edge detail preservation in image fusion,infrared and visible images fusion algorithm based on non-subsampled shear-wave transform(NSST)and convolutional neural network image fusion framework(IFCNN)is proposed.First,infrared and visible images are decomposed by NSST.Then,in order to make the low-frequency sub-band image better highlight the contour information,the image is fused using similarity matching fusion rule;for the high-frequency sub-band images,the feature layers are extracted using IFCNN,and the maximum weight image of feature layer can be obtained through L2 regularization,convolution operation,and maximum selection strategy processing,and the high frequency fusion rule can be determined according to the maximum weight image.Finally,the NSST inverse transform is used to obtain the final fusion image.The experimental results show that the proposed algorithm retains the details of image edges and textures,reduces artifacts and noises,and has good visual effects.
作者
杨艳春
高晓宇
党建武
王阳萍
Yang Yanchun;Gao Xiaoyu;Dang Jianwu;Wang Yangping(School of Electronic and Information Engineering,Lanzhou Jiaotong University,Lanzhou,Gansu 730070,China)
出处
《激光与光电子学进展》
CSCD
北大核心
2021年第20期110-118,共9页
Laser & Optoelectronics Progress
基金
长江学者和创新团队发展计划资助(IRT_16R36)
国家自然科学基金(62067006,61562057)
甘肃省科技计划(18JR3RA104)
甘肃省高等学校产业支撑计划(2020C-19)
兰州市科技计划(2019-4-49)
兰州交通大学天佑创新团队项目(TY202003)
兰州交通大学—天津大学联合创新基金(2021052)。
关键词
图像处理
非下采样剪切波变换
红外与可见光图像
卷积神经网络
图像融合
image processing
non-subsampled shearlet transform
infrared and visible images
convolutional neural network
image fusion
