融合分数阶全变分和重叠组稀疏的遥感图像复原算法

Remote Sensing Image Restoration Algorithm Combining Fractional-Order Total Variation and Overlapping Group Sparse

摘要: 脉冲噪声的随机性和高对比度导致其在遥感图像中难以预测和定位，为了去除脉冲噪声，本文提出了一种融合分数阶全变分先验和重叠组稀疏先验的遥感图像复原算法。该模型采用l₀范数作为数据保真项以避免l₁范数的过度惩罚，利用重叠组稀疏先验来消除阶梯效应，同时分数阶全变分先验能够更有效地保留图像中的边缘和纹理信息。我们使用优化最小化算法和交替方向乘子法来进行求解，并与L0-OGSTV、HNHOTV、L0-TV三种算法进行对比，实验结果表明，本文所提出的算法在峰值信噪比和结构相似度上均优于其他几种算法。

Abstract: The randomness and high contrast of impulse noise cause it to be difficult to predict and localize in remote sensing images. In order to remove the impulse noise, this paper proposes a remote sensing image restoration algorithm that integrates fractional-order total variational prior and overlapping group sparse prior. The model adopts the l₀ norm as the data fidelity term to avoid the over-penalization of the l₁ norm, and utilizes the overlapping group sparse prior to eliminating the staircase effect, while the fractional-order total variation prior can retain the edge and texture information in the image more effectively. We use the majorization-minimization algorithm and the alternating direction multiplier method to solve the problem and compare it with the three algorithms, L0-OGSTV, HNHOTV, and L0-TV, and the experimental results show that the algorithm proposed in this paper outperforms the other algorithms in terms of the peak signal-to-noise ratio and the structural similarity.