Cyclic Weighted Median Method for L1 Low-Rank with Nonlocal for Single color image denoising

A challenging problem in computer vision is how to restore clean image from the noisy one. Low rank decomposition based on Li norm has become a popular solution. However, Since the matrix factorization is nonconvex and Li norm is non-smooth, most methods cannot be truly realized and only suboptimal...

Full description

Saved in:
Bibliographic Details
Published in2020 10th Institute of Electrical and Electronics Engineers International Conference on Cyber Technology in Automation, Control, and Intelligent Systems (CYBER) pp. 370 - 374
Main Authors Luo, Qiong, Liu, Jiaxin, Han, Zhi, Li, Yong, Tang, Yandong, Li, Shengchuan
Format Conference Proceeding
LanguageEnglish
Published IEEE 10.10.2020
Subjects
Complexity theory
Computer vision
Convergence
Gaussian distribution
Minimization
Noise reduction
Optimization
Online AccessGet full text

Cover

More Information
Summary:A challenging problem in computer vision is how to restore clean image from the noisy one. Low rank decomposition based on Li norm has become a popular solution. However, Since the matrix factorization is nonconvex and Li norm is non-smooth, most methods cannot be truly realized and only suboptimal results can be obtained. The cyclic weighted median method alleviates this problem to a certain extent by solving a series of scalars minimization convex sub-problems. However, this method currently can only rely on similar image sequences to find low-rank subspaces, which is seriously degraded in single image denoising. In this paper, we introduce non-local self-similar priors, and apply the cyclic weighted median method for single image denoising for the first time. Experiments prove that our method is better than all competition methods.
DOI:10.1109/CYBER50695.2020.9278965