非下采样Contourlet变换域多聚焦图像融合方法

Novel multi-focus image fusion method in nonsubsampled Contourlet transform domain

针对同一场景的多聚焦图像融合问题,提出基于脉冲耦合神经网络(PCNN)的非下采样Contourlet变换(NSCT)域融合方法.将源图像经过NSCT变换生成的低通子带系数和带通方向子带系数输入PCNN,将各神经元迭代产生的点火频数构成点火映射图.采用接近度函数描述点火映射图邻域特性的关联程度,根据邻域接近度为融合图像选择相应的子带系数,通过NSCT逆变换得到融合结果.实验分析表明,新的融合方法在很大程度上保留了多聚焦图像的清晰区域和特征信息,具有比经典小波变换、Contourlet变换和常规NSCT方法更好的融合性能.

A fusion method using pulse-coupled neural network（PCNN） in non-subsampled Contourlet transform（NSCT） domain was proposed in order to solve the problem of multi-focus image fusion in the same scene.Both the low-pass sub-band coefficient and the band-pass directional sub-band coefficient of source image by NSCT were inputted into PCNN.The ignition mapping image was obtained via the ignition frequency generated by the neuron iteration.Then the approach degree function was adopted to describe the association degree of the neighborhood characteristic in ignition mapping image,and the appropriate sub-band coefficient was selected according to the neighbor approach degree.The fused results were obtained through the inverse NSCT.Experimental results demonstrate that the method greatly retains the clear region and the feature information of multi-focus image.The method has better fusion performance than the classical wavelet transform,the Contourlet transform and the conventional NSCT.