摘要
提出一种新的改进核维纳滤波器图像去噪算法。维纳滤波器通过最小化去噪图像与原始无噪图像之间的均方误差准则进行线性变换完成去噪,维纳滤波器给出了贝叶斯意义下最好的解。但是单纯的线性变换通常难以满足非平稳随机过程,采用核方法将维纳滤波器扩展到特征空间进行非线性变换是一种很好的处理手段。在进行特征空间的非线性维纳滤波时,重构输入空间的原图像时常常会遇到局部收敛性、对初始点较为敏感等问题,考虑到在高斯核函数下,特征空间均方误差和输入空间均方误差有一定的内在相关性,将输入空间下的维纳滤波结果约束到特征空间的均方误差中,在有训练样本下,不但可以进一步提高收敛速度和最优化,而且能更好地恢复原始信号信息。实验结果表明,所提方法的要比传统的维纳滤波和核维纳滤波更为有效。
An improved kernel wiener filter for image denoising is proposed.Image denoising is done by wiener filter using linear transformation,which is minimizing the mean square error between the denoised image and original noiseless image.And the wiener filter can give the optimal solution in the means of Bayesian.But the linear transformation can′t meet the non-stationary process.It is a good method wiener filter of nonlinear transformation in feature space using kernel method.Usually,the nonlinear wiener filter in feature space encounters the problem of localminimum and sensitive to initial values,etc.There is internal relatedness between mean square error in feature space and input space in the Guassian kernel.So here the result of wiener filter in input space is constrained to the mean square error of wiener filter in feature space.The speed of convergence and optimal is improved.The signal information is recovered better.The results of denoising with improved kernel wiener filter are superior to traditional wiener filter and kernel wiener filter.
出处
《激光与红外》
CAS
CSCD
北大核心
2010年第5期549-553,共5页
Laser & Infrared
关键词
图像去噪
维纳滤波
核方法
非线性滤波器
输入空间距离约束
image denoising
wiener filter
kernel method
nonlinear filtering
constraint with distance in input space
