Image component analysis based on sparse signal decomposition and its applications to image processing

在这项研究中，提出了一种图像分解成一组具有不同功能的组件。建议的分解是通过学习图像的代表性微观结构来实现的。图像被分解成组件;每个组件都是由平移的微结构的线性组合生成的。在学习中，稀疏先验被施加在微结构的出现上。将代表结构的学习应用于图像修复，即从含有缺失像素的退化图像中恢复原始图像。通过这种学习方法，从降质图像中成功地学习到了原始图像的微观结构。同时，使用学习的结构恢复丢失的像素。

In this study, an image decomposition into a set of components with different features is proposed. The proposed decomposition is achieved with learning of representative micro structures of an image. The image is decomposed into components ; each of them is generated by a linear combination of the translated microstructures. In the learning, the sparsity prior is imposed on the occurrences of the microstructures. The learning of the representative structure is applied to image inpainting, which is recovery of the original image from a degraded image with missing pixels. By using the learning, the micro structures of the original image are successfully learnt from the degraded image. Simultaneously, the missing pixels are recovered with the learnt structures.