摘要
图像去噪是图像处理中的关键问题之一,也是图像后续处理的基础,结合近年来兴起的稀疏表示理论,能更好地处理图像去噪问题。通过引入图像稀疏表示框架,从含噪图像自身中优化训练字典,初始字典选择构造非采样小波字典来更好地捕获图像信息,通过反复迭代学习获得高度自适应的过完备稀疏字典,重构图像时构造先验概率矩阵,结合后验概率估计与传统的正交匹配算法提出改进的图像重构算法。实验结果表明,与其他去噪方法相比,该算法具有良好的去噪能力,能较好地保持图像的边缘和细节特征,去噪后的图像更为清晰。
Image denoising is one of the key issues in image processing as well as the foundation of image post-processing. To combine the sparse representation theory arisen in recent years can process image denoising better. Through introducing the framework of image sparse representation,we optimise the training dictionary from the noisy image itself. For initial dictionary,we choose undecimated wavelet dictionary to better acquire the information of image. The highly adaptive overcomplete sparse dictionary can be acquired through repeated iteration learning. When reconstructing image,we establish priori probability matrix and propose the improved image reconstruction algorithm by combining the posteriori probability estimation with traditional orthogonal matching algorithm. Experimental result shows that compared with other denoising methods,this algorithm has good denoising ability and can well keep edges and detailed features of image. What＇s more,the image gets clearer after denoising.
出处
《计算机应用与软件》
CSCD
2015年第12期193-196,205,共5页
Computer Applications and Software
基金
江苏省教育厅项目(12KJB520001)
关键词
图像去噪
稀疏表示
非抽样小波
过完备字典
Image denoising
Sparse representation
Undecimated wavelet
Overcomplete dictionary
