Tensor methods for low-level vision

摘要: This chapter introduces tensor methods that can be used for low-level vision tasks such as inpainting, denoising, deblurring, and superresolution. The tensor modeling approaches can be divided into two parts: penalty function minimization and tensor decomposition. We show that many models for penalty function minimization can be optimized by the framework of proximal splitting, and many models for tensor decomposition can be optimized by the framework of majorization-minimization. In addition, we will briefly introduce the higher-order tensor representations of images and show the results of simple experiments.