This paper introduces a new spatially adaptive directional total generalized variation (TGV) regularization for image denoising. Since pixel-specific estimates are not assigned to the minor semi-axis and angle parameters in directional TGV structure, this regularization scheme can be mainly used for images whose texture follows one direction. In our proposed spatially adaptive regularization technique, we develop the directional TGV regularization to a spatially adaptive directional TGV regularization model, which can work efficiently for images whose texture is composed of different directions. Computationally, we adopt a primal-dual algorithm to obtain the optimal solution. Quantitative and qualitative assessments of the new model distinctly demonstrate its competitive performance in feature preservation and staircasing effect suppression.

本文介绍了一种新的用于图像去噪的空间自适应方向总广义变化 (TGV) 正则化。由于在定向 TGV 结构中未将特定像素估计分配给次半轴和角度参数，因此该正则化方案可主要用于纹理遵循一个方向的图像。在我们提出的空间自适应正则化技术中，我们将定向 TGV 正则化发展为空间自适应定向 TGV 正则化模型，该模型可以有效地处理纹理由不同方向组成的图像。在计算上，我们采用原始对偶算法来获得最优解。新模型的定量和定性评估清楚地表明了其在特征保存和阶梯效应抑制方面的竞争性能。