English

应用科学学报

• 论文 • 上一篇    下一篇

Curvelet变换域自适应收缩图像去噪

邓承志1 曹汉强1 汪胜前2   

  1. 1. 华中科技大学 电子与信息工程系，湖北 武汉 430074；
    2. 江西科技师范学院 江西省光电子与通信重点实验室，江西 南昌 330013
  • 收稿日期:2007-05-13 修回日期:2007-08-29 出版日期:2008-01-31 发布日期:2008-01-31

Adaptive Shrinkage Image Denoising in Curvelet Domain

DENG Cheng-zhi 1, CAO Han-qiang 1, WANG Sheng-qian 2   

  1. 1. Department of Electronic & Information, Huazhong University of Science & Technology, Wuhan 430074, China;
    2. Key Lab of Optic-Electronic & Communication, Jiangxi Science & Technology Normal College, Nanchang 330013, China
  • Received:2007-05-13 Revised:2007-08-29 Online:2008-01-31 Published:2008-01-31

被引次数

13

摘要： 研究了Curvelet变换域非参数贝叶斯估计图像去噪问题。利用先验概率模型－正态反高斯(NIG)分布对图像Curvelet系数的稀疏分布进行统计建模，并在此基础上设计出基于NIG的最大后验概率(MAP)估计器。通过估计Curvelet子带系数分布的参数，实现基于MAP的子带自适应收缩图像去噪，最后通过仿真验证了去噪算法的性能。结果表明，该方法能有效地去除图像中的噪声，同时较好地保留了图像的纹理和边缘等细节。

关键词: Curvelet变换, 正态反高斯分布, 最大后验概率(MAP), 图像去噪

Abstract: A nonparametric Bayesian estimator for image denoising in the curvelet domain is studied. A prior model, named normal inverse Gaussian (NIG), is imposed on the curvelet coefficients designed to capture the sparseness of the curvelet expansion. Based on this, a NIG-based maximum a posteriori (MAP) estimator is designed. By estimating the model parameters of curvelet subband coefficients, a MAP-based subband adaptive shrinkage image denoising is realized. Simulation is carried out to show effectiveness of the denoiser. Experimental results show that the proposed method can effectively reduce noise while keep details.

Key words: curvelet transform, normal inverse Gaussian (NIG), maximum a posteriori, image denoising