摘要
提出一种基于Bandelet变换的图像去噪方法,以提高高噪声方差的图像去噪效果。Bandelet变换的核心是Lagrangian函数代价项的准确选取,本文从图像基追踪稀疏模型表示原理和图像阈值去噪方法的内在关系入手,重新定义Lagrangian函数,从而使图像稀疏去噪模型含义更明确,计算更简单。在去噪过程中,首先采用二维平移不变小波变换把图像分解为高频子带;然后用局部Bandelet块估计Bayes阈值确定Lagrangian函数的代价因子,从而对各个高频实施Bandelet化;最后对高频图像系数Bayes软阈值收缩实现图像去噪。国际标准中几何特征明显图像测试表明:在高斯白噪声的方差低于502时,本文方法的去噪效果和目前最好方法的效果相当;当噪声的方差等于或者高于502时,本文去噪方法效果更好。
To improve the efficiency of image denoising at the high Gaussian white noise level,a novel scheme is proposed by combining with the second redundant Bandelet transform version based on the Sparse Land-Basis Pursuit.Starting from the interrelation between Basis Pursuit denoising and threshold denoising with the shrinkage method,the Lagrangian cost function is renewed to minimize the influence of noises and have clearer meanings,and also lead to reduction of computation complexity.There are three steps in the image denoising process.Firstly,translation of invariant 2D wavelets is used to obtain the redundant Bandelet transform version.Secondly,during finding the best geometrical flow and optimal quadtree segments,the cost term of Lagrangian is confirmed by the Bayes estimator.Thirdly,Bayes soft-threshold shrinkage denoising in the bandelet transform domain is implemented.This leads to the state-of-the-art denoising performance,equivalent and sometimes surpassing recently published leading alternative denoising methods,especially as the noise variance is equal to and larger than 502.
出处
《铁道学报》
EI
CSCD
北大核心
2010年第5期108-113,共6页
Journal of the China Railway Society
