The multiwavelet transform has properties of orthogonality, tight frame, and symmetry, which are vital to signal processing. In this study, a new transform, called as MNSDFB, is proposed by combining the multi⁃wavelet and nonsubsampled directional filter banks. The domain transform recursive filter is adopted to fuse the filters after the spatial frequency of the high frequency coefficient is calculated. The modified sum⁃modified⁃Laplacian ( SML) is employed in the low pass subbands as a focus measurement to select fused coefficients. The presented fusion rule in the high pass subband can distinguish the focused regions from the blurred regions. The proposed fusion method was compared with three other fusion methods. The experimental results demonstrate that the proposed fusion meth⁃od can select the focused regions while introducing few artifacts into the final merged image. Furthermore, its objec⁃tive criteria, such as MI and QAB/F, are better than those of the other three methods.%多小波能同时满足正交、紧支、对称等对信号处理十分重要的特性，结合多小波变换的多尺度特点和非子采样方向滤波器组变换的多方向性，提出了一种新的基于多小波和非子采样方向滤波器组的多尺度多方向变换。对于高频系数首先计算其修改空间频率，然后利用域变换递归滤波进行滤波的融合规则；低频系数采用了修改拉普拉斯能量和的（ SML）融合规则。通过与其他融合方法进行实验对比，实验结果表明：本文提出的融合方法能够更加有效地选择源图像中的聚焦良好区域，并且引入的伪影信息较少；此外，与其他融合方法相比本文方法的客观评价结果也是最好的。
展开▼
