基于梯度域导向滤波器和改进PCNN的图像融合算法  

Image fusion algorithm based on gradient domain guided filtering and improved PCNN

作　　者：王健[1,2] 何自豪 刘洁[1] 杨珂[1] WANG Jian;HE Zihao;LIU Jie;YANG Ke(Electronic and Information College,Northwestern Polytechnical University,Xi’an 710129,China;No.365 Institute,Northwestern Polytechnical University,Xi’an 710065,China)

机构地区：[1]西北工业大学电子与信息学院,陕西西安710129 [2]西北工业大学第365所,陕西西安710065

出　　处：《系统工程与电子技术》2022年第8期2381-2392,共12页Systems Engineering and Electronics

基　　金：国家自然科学基金(61671383);陕西省重点产业创新链项目(2018ZDCXL-G-12-2,2019ZDLGY14-02-02,2019ZDLGY14-02-03);空天地海一体化大数据应用技术国家工程实验室;西北工业大学研究生创新基金(Z2017144)资助课题。

摘　　要：为解决当前融合后图像存在的光晕伪影现象以及不利于视觉感知的问题,提出了一种基于梯度域导向滤波(gradient domain guided filtering, GDGF)和改进的脉冲耦合神经网络(pulse-coupled neural network, PCNN)的图像融合算法。首先,利用图像结构、清晰度以及对比度显著性的图像特征构建图像融合模型。其次,采用梯度域导向滤波取代传统优化方法,通过像素间相关性优化初始决策图。然后,将优化决策图作为外部输入刺激改进PCNN模型,得到融合权重图。最后,对源图像和融合权重图进行加权操作得到最终融合图像。实验结果表明,所提方法更好地保留图像边缘、纹理和细节信息,避免目标边缘的光晕伪影现象,且利于视觉观察。In order to solve the problems of halo artefacts and unfavourable visual perception in the fused images, this paper proposes an image fusion algorithm based on gradient-domain guided filtering and an improved pulse-coupled neural network(PCNN). First, an image fusion model is constructed using the image features of image structure, sharpness and contrast saliency. Secondly, the initial decision map is optimised by inter-pixel correlation using gradient-domain guided filtering instead of the traditional optimisation method. Then, the optimised decision map is used as external input to stimulate the improved PCNN model to obtain the fusion weight map. Finally, the source image and the fusion weight map are weighted to obtain the final Finally, the source image and the fusion weight map are weighted to obtain the final fused image. The experimental results show that this method can better preserve the image edge, texture and detail information, avoid halo artefacts on the target edge, and facilitate visual observation.

关 键 词：导向滤波器 改进的脉冲耦合神经网络 图像融合 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]