摘要
针对总变分(TV)模型在去噪时会产生阶梯效应、纹理细节信息丢失的问题,提出一种新的乘性噪声去除模型。该模型采用分数阶变分(FV)和二阶总广义变分(TGV)的正则项。FV的正则项在去噪时可保留图像的纹理信息,TGV的正则项在去噪时可改善阶梯效应。为有效求解该模型,先采用分裂法和交替方向法将原问题转化为两个相关的子问题,再利用原始对偶和梯度下降算法分别对子问题进行求解。对仿真图像进行实验的结果表明,新模型不仅可以提高图像的信噪比,而且可以更好地保留图像的边缘和纹理细节信息,改善阶梯效应。
In this paper, we proposed a novel model based on fraction-order variation (FV) and second-order total generalized variation (TGV), to remove multiplicative noise. In the new model, FV could keep the textures well, and TGV could make the edges well and avoid the staircase effect. In order to solve this model, we used the splitting method and alternating direction method to divide the primal problem into two relevant subproblems at first, and then used the gradient descent method and primal-dual method to solve the subproblem respectively.The numerical experimental results show that the proposed model can not only remove noise effectively but also preserve edge, keep the texture and other details well. At the same time, this model can suppress the staircase effect and improve the signal-to-noise ratio of the image.
出处
《地理空间信息》
2019年第7期126-130,I0003,共6页
Geospatial Information
关键词
乘性噪声
FV
TGV
阶梯效应
原始对偶
multiplicative noise
FV
TGV
staircase effect
primal-dual method
