提出一种Tetrolet框架下基于联合稀疏表示结合改进脉冲耦合神经网络规则的红外与可见光图像融合方法.对源红外与可见光图像进行不考虑旋转和反射情况下的Tetrolet系数分解; 采用联合稀疏方法进行低频系数融合, 通过学习字典进行低频系数的精确拟合并融合.在高频子带系数融合上, 采用改进脉冲耦合神经网络设置相应的融合规则, 根据神经元的点火次数来选择融合图像的高频系数; 并对处理后的高低频系数值进行Tetrolet逆变换获取最终融合结果.结果表明, 该方法能够有效保留待融合图像的边缘与细节特征, 融合结果具有良好的视觉效果, 能够增强观察者对于场景的感知和重要目标的识别能力.在互信息、梯度信息、结构相似度以及视觉敏感度指标上都优于传统变换域融合方法, 尤其在结构相似度以及梯度保持度上分别领先0.033和0.025, 具有有效性.

An fusion method based on joint sparse representation and improved pulse coupled neural network in tetrolet framework was proposed. The original infrared and visible images were decomposed without considering the rotation and reflection; for the low-frequency sub-band coefficients, the joint sparse representation method was used to accurately fit and fuse the low-frequency coefficients through the learning dictionary. In the high frequency subband coefficient fusion, the corresponding fusion rule was set by using the improved pulse coupled neural network, and the high frequency coefficient of the fused image was selected according to the number of firings of the neuron. The processed coefficient values were inversely transformed by tetrolet frame to obtain the final fusion result. The results show that the proposed method can effectively preserve the edge and detail features of the image to be fused, and the fusion results have better visual effects, which can enhance the observer′s ability to perceive the scene and identify important targets. It is superior to the traditional transform domain fusion method in mutual information, gradient information, structural similarity and visual sensitivity index, especially in terms of structural similarity and gradient retention, leading by 0.033 and 0.025, respectively, and has effectiveness.