基于小波变换和多通道脉冲耦合神经网络的高光谱图像融合

Fusion algorithm of hyperspectral images based on wavelet transform and multi-channel PCNN

针对高光谱多波段图像融合的问题,提出了一种基于小波变换和多通道脉冲耦合神经网络模型的新融合方法。该算法利用小波变换对图像进行多尺度分解,将得到的低频和高频系数分别采用多通道PCNN模型进行非线性融合处理,对低频子带直接利用其点火频率图得到融合结果,对各高频子带则利用点火频率图的直方图矢量重心及偏差计算自适应阈值并进行区域分割,对不同的区域采用不同的融合规则进行融合处理;最后进行小波重构得到融合图像。对OMIS高光谱图像的实验结果表明:所提方法能够有效地融合高光谱多个波段图像信息,且纹理细节信息突出。

A novel fusion algorithm based on Pulse Coupled Neural Networks （PCNN） is proposed to fuse multi-band of hyperspectral images. Using wavelet transform to decompose images into multi- levels for extracting the approximation and high frequency information in different orientations. Each sub-band in wavelet domain of multiple images is input to a multi-channel PCNN model to nonlinear fusion. Then the fused approximation sub-band is generated directly using the firing map. Various high frequency sub-bands are segmented into different regions by the histogram vector center of gravity and deviation of the firing matrix and are fused by different fusion rules in different regions. The fusion image is reconstructed by inverse wavelet transform. Experiment results of OMIS images show that the proposed algorithm can effectively fuse multi-band of hyperspectral images and derive rich information of textures and details.