基于改进引导滤波与DCPCNN的图像融合方法

针对红外图像与可见光图像融合易产生边缘信息缺失，目标不够突出等问题，将引导滤波的保持边缘特性与双通道脉冲耦合神经网络（DCPCNN）的非线性耦合调制特性相结合，提出一种基于改进引导滤波与DCPCNN的红外与可见光图像融合算法。该方法首先选取非下采样剪切波变换将图像进行分解，获得高低频分量；对低频分量的融合是利用改进空间频率作用DCPCNN输入激励，且其链接强度由表征图像信息的平均梯度自适应调整来确定；高频分量融合是利用局部平均梯度与区域方差自适应加权，而后采用改进的引导滤波进行平滑处理实现空间一致性；最后，对分别处理后的各分量经过非下采样剪切波变换可逆变换获取融合图像。针对典型背景目标和复杂背景目标两类实验结果表明，与经典的曲波变换、双树复小波变换、非下采样轮廓波变换和非下采样剪切波变换等方法相比，该算法可以有效综合图像的优势信息，且在平均梯度、标准差、空间频率、相关系数等方面具有更高的优势。

Infrared and Visual Image Fusion Method Based on Improved Guided Filtering and Dual-Channel PCNN

Aiming at the infrared images and visible images fusion is easy to produce edge information missing and the target is not enough, this paper combines the edge-preserving characteristics of the guided filtering with the non-linear modulation characteristics of the Dual-Channel Pulse Coupled Neural Network（DCPCNN）, an infrared and visible image fusion algorithm based on improved guided filtering and DCPCNN is proposed. Firstly, this paper obtains the high and low frequency components by using the Non-Subsampled Shearlet Transform（NSST） multi-scale decomposition of the strictly registered source images. Secondly, the low frequency components of the image is obtained by using the modified spatial frequency as the external excitation of the dual-channel PCNN, at the same time, the average gradient of low frequency components are used to adjust the link strength adaptively. Moreover, for the high frequency components of the image fusion, this paper uses local average gradient and regional variance adaptive weighting, and then uses improved guided filtering for smoothing to achieve spatial consistency. Finally, the paper uses the NSST inverse transform method to fuse low and high frequency components to obtain a fused image. Two types of experimental results for typical background targets and complex background targets show that proposed method can effectively synthesize the superiority information of the image. The proposed method is superior to the methods based on classical curvelet transform, dual-tree complex wavelet transform, Non-Subsampled Contourlet Transform（NSCT） and NSST in the quantitative evaluation indexes such as average gradient, standard deviation, spatial frequency and correlation coefficient.