基于方向滤波器组与Laplacian能量和的图像融合算法

Image Fusion Algorithm Based on Direction Filter Bank and Laplacian Energy Sum

中文关键词: 图像融合  多NSDFB  局部改进的Laplacian能量和  脉冲耦合神经网络  非下采样方向滤波器组

英文关键词: image fusion  multi-NSDFB  Local Sum-Modified-Laplacian  pulse couple neural network  non-subsampled direction filter bank

中文摘要:

      目的 针对基于Contourlet变换的融合算法在边缘上易出现吉布斯现象，使其融合图像产生几何失真的问题，设计一种非下采样方向滤波器组耦合局部Laplacian能量和的图像融合算法。方法 首先，结合多小波变换（multi-wavelet transform，MWT）与非下采样方向滤波器组（Non-Subsampled Direction Filter Bank，NSDFB），将图像分解为3个高频方向系数和1个低频系数。对于低频系数，采用局部修正的Laplacian能量和（Local Sum-Modified- Laplacian，LSML）与脉冲耦合神经网络（Pulse couple neural network，PCNN）组合的LSML-PCNN模型来完成低频信息的融合。对于高频系数，通过提取低频和高频子带边缘，并利用系数绝对最大值法作为依据，实现高频系数的融合。结果 实验数据表明，与当前图像融合方案相比，所提算法具有更高的融合质量，得到的融合图像边缘更加清晰和完整。结论 所提算法拥有较高的融合视觉效果，可改善图像的对比度和分辨率，在图像处理领域具有一定的参考价值。

英文摘要:

      The work aims to propose an image fusion algorithm based on Non-Subsampled Direction Filter Bank (NSDFB) coupling local Laplacian energy sum, regarding the fusion algorithm based on the Contourlet transform that easily leads to the Gibbs phenomenon on the edge. Firstly, combined with multi-wavelet transform (MWT) and NSDFB, the image was decomposed into three high frequency directional coefficients and one low frequency coefficient. Then, for low frequency coefficient, the LSML-PCNN model composed of Local Sum-Modified-Laplacian (LSML) and pulse couple neural network (PCNN) was used to fuse the low frequency information. For the high frequency coefficient, the low frequency and high frequency subband edges were extracted, and the absolute maximum value method of coefficient was used as the basis to achieve the fusion of high frequency coefficients. The experimental results showed that, the proposed algorithm had higher fusion quality and clearer and more complete edges of the fusion image than the current image fusion scheme. The proposed algorithm has a higher fusion visual effect and it can improve the contrast and resolution of images, which has certain reference value in the field of image processing.