Image fusion by combining multiwavelet with nonsubsampled direction filter bank
Image fusion by combining multiwavelet with nonsubsampled direction filter bank
复制标题
DOI：
10.1007/s00500-015-1893-0
复制
发表时间：
2015-12
期刊：
影响因子：
4.1
通讯作者：
Geng Peng;Zhengyou Wang;Shuaiqi Liu-;Shanna Zhuang
中科院分区：
文献类型：
--
作者：
Geng Peng;Zhengyou Wang;Shuaiqi Liu-;Shanna Zhuang
针对非下采样轮廓波变换和剪切波变换降阶过大的问题，将多小波变换与非下采样方向滤波器组相结合，提出了一种新的变换方法。随后，提出了一种基于多尺度分解的图像融合方法。脉冲耦合神经网络(PCNN)是由每个子带系数的局部和修正拉普拉斯测量激励的。如果这些系数比其他系数产生更大的发射时间，则将选择这些系数来合成融合图像。分别对多聚焦图像、医学图像和多光谱图像进行了实验。实验表明，该融合方法明显优于传统的小波、NSCT、Shearlet等基于PCNN的多尺度几何分析方法。
Aiming to solving the problem of too much reductancy in nonsubsampled contourlet transform and shearlet transform, a new type of transform by combining the multiwavelet transform with nonsubsampled direction filter bank is proposed. Subsequently, a multi-scale-decomposition-based image fusion approach is presented. The pulse coupled neural networks (PCNN) are motivated by the local sum-modified-Laplacian measurement of every subband coefficient. If the coefficients generate larger firing times than the other, the coefficients will be chose to synthesize the fused image. Several experiments are performed on three kinds of images, such as multi-focus images, medical images and multispectral images. The experiments indicate that the proposed fusion method observably outperforms the other multi-scale geometry analysis methods adopting the PCNN, such as the traditional wavelet, NSCT, shearlet and other latest image fusion algorithms.