基于分水岭和广义非局部平均的小波去噪  

Wavelet denoising based on watershed and generalized non-local means

作　　者：李万臣 葛磊[1] 

机构地区：[1]哈尔滨工程大学信息与通信工程学院,黑龙江哈尔滨150001

出　　处：《应用科技》2011年第4期24-29,共6页Applied Science and Technology

摘　　要：非局部平均滤波去噪方法和基于广义非局部平均的小波域去噪方法都会在不同程度上损失图像细节信息.为了在去除图像噪声的同时更好地保留图像细节,文中提出了一种基于分水岭分割和广义非局部平均的小波去噪方法.首先对含有噪声的图像进行基于梯度的分水岭分割并保留分水岭脊线;然后对含有噪声的图像进行多维度小波分解,对分解的每一层系数估计尺度系数和形状系数,构造每层小波子系数的广义高斯模型,对每层细节子带信息分别在水平、垂直、对角线3个方向应用基于广义高斯模型的非局部平均滤波;最后用含噪图像中与分水岭脊线相对应的像素点替换小波重构后图像的对应像素点.仿真结果表明,该方法与基于广义非局部平局的小波分析去噪法相比能获得更好的视觉效果和去噪效果.Non-local means filter de-noising method and wavelet de-noising method which is based on generalized non-local means filter are both ineffective in reserving image＇s detailed information when using them for image denoising.For this shortcoming,this paper puts forward a wavelet de-noising method based on watershed segmentation and the generalized non-local means filter.Firstly,in order to segment the noised image and keep the ridge,we applied the watershed segmentation based on gradient method;then the noised image was multidimensionally wavelet decomposed,and we evaluated scale coefficient and shape coefficient for every layer of coefficients decomposed,after which the generalized Gaussian model was constructed for each layer of wavelet sub-coefficients,and then apply non-local means filter based on generalized Gaussian model respectively from horizontal,vertical and diagonal directions for sub-band information of every layer details.Finally,the noised image pixels which correspond to the watershed ridge replaced the corresponding pixels in the wavelet reconstructed image.Simulation results showed that this method can obtain better vision effect and denoising effect than the method based on the wavelet analysis of generalized non-local means.

关 键 词：分水岭 小波分析 非局部平均 图像去噪 

分 类 号：TN911.73[电子电信—通信与信息系统]