基于MNLMF和SF方向滤波的图像融合算法

针对非下采样金子塔(NSP)用于平移不变的剪切波变换(NSST)中不能有效捕获图像的结构信息问题,提出一种多尺度非局部均值滤波(MNLMF)和剪切波(SF)方向滤波的新变换,即利用MNLMF代替NSST中NSP分解,然后将其用到图像融合中,并将输入图像分解成不同子带.对于近似子带,采用区域像素能量(PE)与梯度能量(GE)加权和的融合规则;对于方向子带,提出基于GE与系数绝对值(CAV)混合的融合规则.同时提出了基于MNLMF与SF的图像融合算法.仿真对比实验表明,所提出的方法在视觉感知和客观质量评价两个方面具有明显的优势.

Image fusion method based on multi-scale non-local mean filter and shear direction filter

For the problem that the non-subsampled pyraimd(NSP) used in non-subsampled shearlet transform(NSST) cannot effectively capture the image structure information, a multi-scale nonlocal means filter(MNMF) combined directional shearinge filter(SF) of new transformation is proposed, which uses the MNMF instead of the NSP decomposition in the NSST transformation. Then, it is used in image fusion, and the input image is decomposed into different subbands.The approximation subbands are fused by applying the regional weighted sum of pixel energy(PE) and gradient energy(GE). For the direction subbands, the fusion rule based on GE combiend with the coefficient of absolute value(CAV) is proposed. The simulation experiments verify that the proposed fusion method has obvious advantages in two aspects of visual perception and objective quality evaluation.