抑制全变分阶梯效应的新方法

摘要

全变分去噪算法能够很好地保留图像边缘和细节信息,但在图像平滑区域会产生严重的阶梯效应,而邻域线性回归滤波器可以避免产生阶梯效应.针对这一点,提出一种抑制全变分阶梯效应的新方法.该方法利用改进的高斯函数作为权函数,根据图像不同区域的灰度特性,自适应地调节权值,从而将全变分算法与邻域线性回归滤波有效地结合,达到优势互补,既可以保留图像的边缘信息,又可以抑制平滑区域的阶梯效应.比较了上述三种方法的结果表明,新方法具有更好的复原效果.%Total Variation minimization is a successful approach to recover images with sharp edges and details. Unfortunately,it shows a staircase effect. Linear regression neighborhood filter can solve the problem. This paper proposes a new method for avoiding the staircase effect. It combines the total variational filter with linear regression neighborhood filter. The combined technique is able to preserve edges and avoid the staircase effect in smooth regions. A weight function based on improved Gaussain function can be found adaptively depending on the characteristics of the image. The performance of the new method is compared with the two other methods proposed above.Experimental results show that the new method has the better restored effect.