基于交叠组合稀疏全变分的图像去噪方法

全变分作为一种常用的去噪模型,在图像去噪中较好地保持图像边缘信息,但是容易产生"阶梯效应"。为了克服这个缺点,提出一种基于快速傅里叶变换的交叠组合稀疏全变分去噪模型。首先,充分考虑图像梯度的邻域结构相识性,通过交叠组合计算像素点的梯度,以凸显平滑区域的高噪声污染点和边界区域像素点的差异。然后,基于快速傅里叶变换和交替方向乘子算法在频域中求解去噪模型。实验结果表明,新模型在保护图像边缘信息的同时,有效去除噪声,同时抑制"阶梯效应"。与几种较好的去噪算法相比,新模型的峰值信噪比、结构相识度、视觉效果、计算效率均有明显提高。

An Image Denoising Method Based on Overlapping Group Sparsity Total Variation

As one of the common image denoising models, the total variation transform has obvious advantages in preserving image edges, while it may produce some undesired stair-case artifacts. To overcome the drawback mentioned above, an overlapping group sparsity total variation model was proposed for image denoising instead of traditional TV model. Considered the structure similarity of neighborhood image gradient, the image gradient had been calculated by overlapping group algorithm to make a distinction between high noise pollution point in smooth region and pixels in boundary regions of image. Then by introducing fast Fourier transform and alternating direction method of multipliers algorithm, the denoising issue was solved in frequency domain. The experimental results demonstrate that the proposed method can remove image noise while the edge and texture details of the image are well protected, avoiding stair-case artifacts. Compared with the other state-of-the-art algorithms, the proposed method leads to better peak signal noise ratio, structure similarity and visual effects, while the computational efficiency is improved.