基于结构信息和稀疏贝叶斯学习的图像去噪

Image Denoising Based on Structural Information and Sparse Bayesian Learning

基于冗余字典的信号稀疏分解采用超完备的冗余函数系统代替传统的正交基函数,从而为信号自适应地稀疏扩展提供了极大的灵活性。该字典可以高效的实现数据压缩,更重要的是可以利用字典的冗余特性更全面的捕捉原始信号的自然特征。本文采用稀疏贝叶斯学习的方法构造冗余字典,应用图像的结构信息作为荣誉字典的先验,实现图像去噪。该方法可以在图像处理过程中在原位学习字典,并可以自适应的调整所使用字典的大小。另一方面,该方面不需要预先设定噪声值,并且可以应用顺序推断,因此可以用于大规模的图像。实验结果表明,该方法可以很好的实现图像去噪。

Signal sparse decomposition based on redundant dictionary utilizes over- complete redundancy function system to substitute traditional orthogonal basis function so as to provide tremendous flexibility for sparse extending signal adaptively. The dictionary can be used to compress the data with high efficiency; and what is more important is that using the redundant speciality of dictionary can capture the natural characteristics of the original signal more comprehensively. The sparse Bayesian learning method is employed to construct the redundant dictionary,and using structural Information of the image as prior knowledge of credited dictionary to implement image denoising. Using this method can learn dictionary in original place during image processing,and adjust size of the used dictionary adaptively. In other hand,the noise value is not needed to be preset in this method,and it can be deduced sequentially. Therefore it can be used for large scale image. The experiment result shows that using this method can implement image denoising better.