摘要
数据的描述方法对提取数据特征至关重要,通常这种描述方法是基于数据的线性变换。传统的的傅立叶变换、离散余弦变换、主分量分析等线性变换方法都是基于全局变换的思想,无法反映图像在时频域的局部特征。独立分量分析是一种多维数据线性变换的方法,它从数据间的高阶统计特性出发,提取的图像数据特征基函数在空间频域中体现了方向性和局部性,能很好的自适应图像数据,并且其所得系数具有稀疏分布的特性。用它对无噪声图像数据进行学习,利用得到的稀疏码变换矩阵对噪声图像数据进行稀疏码变换,得到稀疏成分,并结合最大似然估计得到的软门限算子对该稀疏成分进行收缩,从而达到了去除图像噪声的目的。试验表明该方法在去噪效果和保存图像细节方面明显优于传统的维纳滤波方法。
Data representations are important for extracting data features,usually the representations are based on the linear transformation of the data. Traditional linear transformations, such as Fourier Transformation, Discret Cosin Transformation and Principal Component Analysis etc, are all based on the entire idea, they can not reflect some local features of image in the field of time and frequency. Independent Component Analysis （ICA） is a multi-dimention data transformation method. Rased on the high-order statistics of data,the feature basis function which is obtained by ICA from image data are localized in space as well as in frequency and orientation ,fit image data well. Besides, the obtained components have a sparse distribution. So first, ICA is used to estimate the sparse code transformation from noise-free image data. Then ,this transformation is used to transform the noisy image data into sparse components. Finally, the soft threshold which is got by Maximum Likelihood Estimation is used to shrinkage the sparse components to achieve denoising. The results show that this method have a better effect in denoising and protecting image details,compared to the traditional method of Wiener filter.
出处
《信号处理》
CSCD
北大核心
2007年第5期742-746,共5页
Journal of Signal Processing
关键词
稀疏码收缩
独立分量分析
软门限
图像去噪
Sparse Code Shrinkage
Independent Component Analysis（ICA）
Soft Threshold
Image-denoising
