方向邻域全变分图像去噪

为了弥补传统全变分（TV）算法忽略了图像边缘方向的不足, 结合梯度幅度和方向提出了基于方向全变分的去噪算法。该算法运用图像梯度幅度将图像像素划分为边缘区域和非边缘区域, 运用梯度方向对不同区域的像素选取不同的四邻域像素, 针对不同邻域对传统TV算法进行离散分析, 完成了图像的保边去噪。实验结果表明, 结合边缘方向信息改进了传统TV算法的邻域选择方式, 不仅更好地保留了图像边缘信息和重要细节, 且提高了图像的PSNR和视觉效果。

Image denoising on direction total variation

Traditional algorithms of total variation often neglect the images' edge direction. To remedy this defect, this paper proposed a new method of image denoising on direction total variation, combining gradient magnitude and orientation. It divided the image into edge regions and non-edge regions on edge gradient magnitude, and chose different four-neighborhood for different regions based on gradient magnitude, to various of the neighborhoods using analysis of variance. So it can preserves image edges during image denoising. Experimental results show that the proposed method combining edge direction information can improve the neighborhood selection strategy in traditional methods of total variation. It not only preserves edges and key details effectively, but also improves the PSNR and vision of denoising images.