光子学报, 2019, 48 (9): 0910003, 网络出版: 2019-10-12
基于高斯曲率优化和非下采样剪切波变换的高密度混合噪声去除算法
High Density Mixed Noise Removal Algorithm Based on Gaussian Curvature Optimization and Nonsubsampled Shearlet Transform
图像降噪 高斯曲率优化 非下采样剪切波变换 混合噪声 阈值收缩 Image denoising Gaussian curvature optimization Nonsubsampled shearlet transform Mixed noise Threshold shrinkage
摘要
为提高矿井混合噪声图像的可观测性, 提出了基于高斯曲率优化和非下采样剪切波变换的高密度混合噪声去除算法.使用局部高斯曲率优化混合噪声图像, 抑制椒盐噪声对噪声分布的影响, 使混合噪声分布近似为高斯噪声分布.使用非下采样剪切波变换分解高斯曲率优化图像, 实施自适应硬阈值收缩降噪, 去除混合噪声中的高斯噪声成分.最后, 迭代使用局部高斯曲率优化和非下采样剪切波变换降噪去除残余噪声, 直至输出图像梯度能量满足停止条件.实验表明, 本文算法能够有效地去除高斯噪声和椒盐噪声构成的高密度混合噪声, 且有效抑制了剪切波变换降噪引起的伪吉布斯现象, 有效地降低了矿井图像的噪声.
Abstract
In order to improve the observability of mine images corrupted by mixed noise, a highdensity mixed noise removal algorithm based on Gaussian curvature optimization and nonsubsampled shearlet transform was proposed. The local Gaussian curvature is used to optimize the mixed noise image to suppress the influence of salt & pepper noise on the noise distribution, which makes the mixed noise distribution approximate to a Gaussian noise distribution. Then, the nonsubsampled shearlet transform is used to decompose the image optimized by Gaussian curvature and implement adaptive hard threshold shrinkage to remove the Gaussian noise in the mixed noise. Finally, the local Gaussian curvature optimization and the nonsubsampled shearlet transform are executed iteratively to reduce the residual noise until the output image gradient energy satisfies the stop condition. Experiments show that the proposed algorithm can effectively remove the highdensity mixed noise composed of Gaussian noise and salt and pepper noise, and effectively suppress the PseudoGibbs phenomenon caused by shearlet transform denoising algorithms, and effectively reduce the noise of mine images.
王满利, 田子建, 桂伟峰, 吴君. 基于高斯曲率优化和非下采样剪切波变换的高密度混合噪声去除算法[J]. 光子学报, 2019, 48(9): 0910003. WANG Manli, TIAN Zijian, GUI Weifeng, WU Jun. High Density Mixed Noise Removal Algorithm Based on Gaussian Curvature Optimization and Nonsubsampled Shearlet Transform[J]. ACTA PHOTONICA SINICA, 2019, 48(9): 0910003.