滤除图像中混合噪声的LSE模型  

The LSE model to denoise mixed noise in images

作　　者：袁珍[1] 林相波[1] 王新宁[1] 

机构地区：[1]大连理工大学信息与通信工程学院,辽宁大连116024

出　　处：《信号处理》2013年第10期1329-1335,共7页Journal of Signal Processing

基　　金：国家自然科学基金资助(61101230);国家科技支撑计划子课题(2013BAK02B04-05)

摘　　要：图像中的高斯白噪声使LS模型中的低秩矩阵低秩性和稀疏矩阵稀疏性不能同时满足,造成去噪不充分或细节严重丢失。本文在LS模型的基础上引入高斯噪声约束项,提出一种新的用于去除图像中混合噪声的LSE模型,该模型首先对图像进行相似块匹配,然后对得到的相似块低秩逼近得到去噪图像。实验结果表明,与LS模型相比,LSE模型在保证去噪效果的同时,保留了图像的细节信息,具有更佳的视觉效果,去噪图像的信噪比提高了约0.1-2dB;与BM3D相比,在高斯噪声较小的情况下信噪比提高了约0.5-2.5dB。The Gaussian white noises in images degrade the LS model,because it can not satisfy low rank property of low rank matrix and the sparse property of sparse matrix.To overcome the disadvantages of LS model,this paper proposed a new model which added the Gaussian restraint to the LS model,named LSE model,to remove random impulse noise and Gaussian white noise in images simultaneously.The experimental results show that compared with LS model,the LSE model can guarantee visual effect and keep the details.The PSNR of the denoised image improved about 0.1-2dB compared with LS model.In the case of Gaussian noise is small,the PSNR of the denoised image improved about 0.5-2.5dB compared with BM3D.

关 键 词：块匹配 低秩理论 增广拉格朗日算法 

分 类 号：TN911.73[电子电信—通信与信息系统]