电子学报 ›› 2021, Vol. 49 ›› Issue (6): 1187-1194.DOI: 10.12263/DZXB.20200606

• 学术论文 • 上一篇    下一篇

基于图像中值灰度相似度函数加权曲率滤波算法

甘建旺1, 沙芸1, 张国英2   

  1. 1. 北京石油化工学院信息工程学院, 北京 102617;
    2. 中国矿业大学(北京)机电与信息工程学院, 北京 100083
  • 收稿日期:2020-06-22 修回日期:2020-09-26 出版日期:2021-06-25
    • 通讯作者:
    • 沙芸
    • 作者简介:
    • 甘建旺 男,1995年出生,湖北黄冈人.现为北京石油化工学院信息工程学院硕士研究生.主要研究方向为图像处理与机器学习.E-mail:2036956253@139.com;张国英 女,1968年出生,北京人,中国矿业大学(北京)机电与信息工程学院教授.主要研究方向为图像处理与机器学习.E-mail:zhangguoying1101@163.com
    • 基金资助:
    • 工业与信息化部2019年工业强基专项

Weighted Curvature Filtering Algorithm Based on Image Median Gray Similarity Function

GAN Jian-wang1, SHA Yun1, ZHANG Guo-ying2   

  1. 1. School of Information Engineering, Beijing Institute of Petrochemical Technology, Beijing 102617, China;
    2. School of Mechanical Electrical & Information Engineering, China University of Mining & Technology, Beijing 100083, China
  • Received:2020-06-22 Revised:2020-09-26 Online:2021-06-25 Published:2022-06-25
    • Corresponding author:
    • SHA Yun
    • Supported by:
    • Ministry of Industry and Information 2019 Industrial Foundation Enhancement Project

摘要: 曲率滤波算法通过构造滤波算子快速优化变分模型,但全变分曲率滤波及高斯曲率滤波易致去噪过平滑且椒盐噪声去除较差.提出了基于图像中值灰度相似度函数加权曲率滤波算法,其中,中值灰度相似度函数方差取决于小波变换最高频子带系数,能较好防止图像过平滑,且提高椒盐噪声去除能力;因此,采用中值灰度相似度函数分别对局部高斯曲率与局部全变分曲率投影算子加权,并分别迭代局部加权高斯曲率投影算子与局部加权全变分曲率投影算子,直至输出图像梯度总能量满足停止条件.实验表明,基于图像中值灰度相似度函数加权全变分曲率滤波与加权高斯曲率滤波比传统全变分曲率滤波和高斯曲率滤波去噪效果更好.

关键词: 加权全变分曲率滤波, 加权高斯曲率滤波, 中值灰度相似度函数, 小波变换, 椒盐噪声

Abstract: The curvature filtering algorithm optimizes the variational model quickly by constructing a filter operator, but the total variational curvature filtering and Gaussian curvature filtering are easy to cause over smooth denoising with poor salt and pepper noise removal. A weighted curvature filtering algorithm based on image median gray similarity function is proposed, in which the variance of the median gray similarity function depends on the highest frequency subband coefficient of wavelet transform, which can prevent the image from being too smooth and improve the ability of removing salt and pepper noise. Therefore, the local Gaussian curvature projection operator and the local total variational curvature projection operator are weighted by the median gray level similarity function, and the local weighted Gaussian curvature projection operator and the local weighted total variational curvature projection operator are iterated respectively until the total energy of the output image gradient meets the stop condition. Experimental results show that the denoising effect of weighted total variation curvature filter and weighted Gaussian curvature filter based on image median gray similarity function is better than the traditional total variation curvature filter and Gaussian curvature filter.

Key words: weighted total variational curvature filter, weighted Gaussian curvature filter, median gray similarity function, wavelet transform, salt and pepper noise

中图分类号: