基于变分贝叶斯推理的高光谱图像恢复
Hyperspectral image restoration based on variational Bayesian inference

提出一种基于变分贝叶斯推理的高光谱图像恢复方法. 建立描述高斯噪声的最大似然函数项，采用小波基矩阵变换，构建小波变换后因子稀疏分布的先验函数；然后建立估计图像和相关超参数的联合后验概率估计模型，并通过变分贝叶斯推理得到估计的图像. 利用实际的高光谱图像进行实验，从恢复的衡量指标和视觉效果图两方面验证所提出方法的有效性，结果优于目前常用的图像恢复方法.
We propose a methed for hyperspectral image restoration based on variational Bayesian methed. One hand，we propose a likelihood function term accounting for Gaussian noise，the other hand，we propose the prior distribution for the sparsity of image in the transformation domain based on the wavelet base. Then we give the maximization posterior model of the image to be estimated and the relevant super-parameter. At last，we infer the maximization posterior function based on variational Bayesian method. From the experiments on the real hyperspectral images，the results show the proposed method can surpass the well known methods，both in terms of indexes and visual effect