Abstract: Aiming at traditional multi-scale fusion algorithm without translation, the fusion effect is poorer, and the parameters setting of Pulse Coupled Neural Network（PCNN） is complex and so on, this paper proposes an image fusion method using the combination NonSubsampled Shearlets Transform（NSST） and Genetic Algorithm（GA） to optimize the parameters of the PCNN, the maximum values of fusion indicators（mutual information [MI], edge information reservation [QAB/F], entropy [EN], spatial frequency [SF], Standard Deviation [STD] and Average Gradient [AG]） are set as the objective function of the GA optimization algorithm to obtain the optimal parameters solution of PCNN, such as linking strength, threshold and so on. Firstly, the source images are decomposed via NSST; then low frequency subband and high frequency subband are fused with Sum-Modified-Laplacian（SML） and PCNN which is guided by Space Frequency（SF）, respectively; finally NSST inverse transformation is used to get the final fusion image. According to the subjective evaluation and objective evaluation indicators, the fused effects of multi-focus image, medical image and infrared and visible light image are carried through the evaluation analysis. The experimental results show that the algorithm objective evaluation indicators are superior to other algorithms, have better fusion effect.

Key words: NonSubsampled Shearlets Transform（NSST）, image fusion, GA optimize PCNN, fusion rules, evaluation indicators

摘要： 针对传统多尺度融合算法不具平移性、融合效果较差以及PCNN参数设置复杂等问题，提出了一种结合非下采样剪切波变换（NSST）与遗传算法（GA）优化脉冲耦合神经网络（PCNN）参数的图像融合方法，将融合指标（互信息[MI]、边缘信息保留度[QAB/F]、熵[EN]、空间频率[SF]、图像标准差[STD]和图像平均梯度[AG]）的最大值设为GA优化算法的目标函数，从而获得最优解对PCNN的链接强度、阈值等参数进行优化。首先利用NSST对图像进行多尺度分解，其次高频采用空间频率引导PCNN进行融合，低频采用改进拉普拉斯能量和（SML）进行融合，最后进行NSST逆变换得到最终的融合图像。根据主观评价与客观评价指标对多聚焦图像、医学图像和红外及可见光图像的融合效果进行评价分析。实验结果表明，该算法在客观评价指标上优于其他算法，有较好的融合效果。

关键词: 非下采样剪切波变换（NSST）, 图像融合, GA优化PCNN, 融合规则, 评价指标