基于NSST和CS的红外与可见光图像融合

针对红外与可见光图像需要实时融合的特点,提出一种降低算法复杂度的基于非降采样剪切波变换(Non-subsampled Shearlet Transform)和压缩感知域的红外与可见光图像融合算法。利用NSST算法对红外图像和可见光图像分别进行多尺度、多方向稀疏分解,分别得到低频系数和各带通方向子带系数。对低频子带系数采用基于目标特征的加权平均融合规则;压缩感知理论的测量矩阵采用哈达马阶快速沃尔什矩阵,对细节信息保留较多的各带通子带系数进行观测测量,得到更稀疏的各带通子带系数测量值,对此测量值采用基于区域方差选大的融合规则得到融合测量值,运用基于增广的拉格朗日乘子和交叠方向恢复算法对融合测量值进行重构得到近似精确的各带通子带融合系数,最后对低频子带融合系数和各带通方向子带融合系数执行NSST逆变换得到最终的融合图像。实验结果表明,该融合方法不仅可以保证融合清晰度,同时还可以缩短算法的运行时间。

Infrared and visible image fusion based on NSST domain and compressed sensing

Aiming at the characteristics of the infrared and visible image fusion,a fusion method that can reduce the algorithm complexity,based on the Non-subsampled Shearlet Transform(NSST)and the Compressed Sensing(CS)is proposed.The source images are decomposed sparsely on multi-direction and multi-scale by using NSST,and then the low-frequency coefficients and every bandpass subband direction coefficients are obtained respectively.For the low frequency subband coefficients,the fusion method of the weighted average based on the target characteristics is adopted;Compressed sensing measurement matrix uses the fast Walsh Hadad matrix.For the bandpass subband coefficients,the fusion method based on the region variance to choose big one is adopted.The measured fusion values are reconstructed by using the recovery algorithm based on the augmented Lagrange multiplier and overlapping direction,and then the approximate precision values of the band pass subband coefficients are achieved.Experimental results show that the fusion method can guarantee the sharpness of the fused image and reduce the running time of the algorithm.