基于脉冲耦合神经网络的异源图像融合方法

针对基于主成分分析与二代小波变换的图像融合算法中鲁棒性不高、融合图像质量较低的问题,提出了基于鲁棒性主成分分析与脉冲耦合神经网络的融合方法。所提出的算法将可见光与红外图像进行二代小波变换,转换为高频与低频信号,接着采用不同的融合策略针对低频和高频信号进行融合。针对低频信号,利用鲁棒性主成分分析法还原低秩矩阵并采用加权平均的融合策略进行融合;针对高频信号,将其送入至脉冲神耦合神经网络中进行融合得到融合后的小波系数。将融合后的小波系数进行逆变换,得到最终融合图像。实验结果表明,相比于基于主成分分析与二代小波变换的图像融合算法,利用所提出的出算法得到的融合图像中熵指标、空间频率指标、结构相似度指标和峰值信噪比指标均得到了不同程度的提升。因此,所提出的算法能够更好地提取目标信息,使融合图像中目标的轮廓边缘更加清晰,同时将提升小波分解出的高频信息利用PCNN进行融合,更加突出细节信息。

Heterogeneous Image Fusion Method Based on Pulse Coupled Neural Network

In order to solve the problems of low robustness and low quality of image fusion algorithm based on principal component analysis and second-generation wavelet transform,a fusion method based on robust principal component analysis and pulsed coupled neural network is proposed.The visible light and infrared images are transformed into high-frequency signal and low-frequency signal by the second-generation wavelet transform,and different fusion strategies will be used for the fusion of low frequency signal and high frequency signal.The low frequency signals are restored to the low-rank matrix by the robust principal component analysis (RPCA) and the weighted average fusion strategy is used for fusion.The high frequency signals are fed into the pulse coupled neural network (PCNN) for fusion.The fused image is obtained by inverting the wavelet coefficients after fusion.The experimental results show that,compared with the image fusion algorithms based on principal component analysis and second-generation wavelet transform,it can be seen that the index of entropy,spatial frequency and structural similarity are improved to varying degrees.The proposed algorithm can better extract target information and make the contour edge of the target in the fusion image clearer.At the same time,the high-frequency information decomposed by lifting wavelet is fused with PCNN to highlight the more detailed information.