基于贝叶斯压缩感知与形态学成分分析的图像修复方法研究

Study of image inpainting method based on Bayesian compressive sensing and morphological component analysis

针对传统图像修复方法依赖图像的结构特征和基于稀疏表示的图像修复方法未考虑修复过程中的观测噪声的问题,提出了一种基于贝叶斯压缩感知与形态学成分分析的图像修复方法。首先通过形态学成分分析法利用Curvelet和局部离散余弦变换分别稀疏图像的结构和纹理部分,然后用贝叶斯压缩感知得到稀疏系数的分布函数,分别求得分布函数的均值和方差,将两个均值作为结构和纹理稀疏系数的估计,方差作为噪声的估计,最后合并两部分的修复结果获得修复后的图像。仿真结果验证该方法可以提高图像的修复质量。

To the question that traditional image inpainting methods depended on the structure characteristics of the image and the image inpainting method based on sparse representation without considering observation noise,this paper introduced an image inpainting method based on Bayesian compressive sensing and morphological component analysis. This algorithm transformed the sparsity for structure and texture with Curvelet and local discrete cosine respectively through morphological component analysis firstly. Then it got the posterior distribution function of the two sparse coefficient through Bayesian compressive sensing. It obtained the mean and the variance of the distribution function. The two means could be used as the estimation of the sparse coefficient for structure and texture of the image,and the variance was the estimation of the noise. Lastly,it obtained the inpainting image after merging the two parts. The emulation results prove that this method can improve the quality of image.