Hyperspectral Image Denoising With Low-rank Constraint

With the development of the remote sensing technology,the research of hyperspectral image(HSI)has raised many people's attention.HSIs have been widely used in many aspects such as food safety,medical imaging,and biometrics.However,for the reason of the hardware limitations and the noise in imaging processing,HSIs are unavoidably polluted by noise.As for HSI denoising,compared with upgrading hardware,a more feasible approach is developing denoising algorithm.Thus,the algorithms for HSI denoising have received extensive research.First,we introduce the development of HSI denoising.After the explanation of the necessity of the denoising algorithm,we introduce the sparse representation and the weighted nuclear norm.Both of them are the basement of the proposed HSI denoising model.The low-rank property is frequently used in image denoising.As an improvement of the nuclear norm,the weighted nuclear norm can provide a better low-rank constraint.Thus,we just utilize the weighted nuclear norm in HSI denoising.Sparse representation is to represent the original signal with as few atoms as possible.Due to the characteristics of the HSIs,sparse representation has received deeply research and it has been widely used in HSI denoising.We find that the HSIs have the strong local low-rank property.And compared to global low rank,local low rank can better utilize the spectral-spatial information of the HSIs.Thus,it is more conducive to HSI denoising.Based on this,we propose a local low-rank and sparse representation model for HSI denoising.Then,we design an algorithm that is under the alternative minimization framework for the proposed model.We numerically show that the convergent history of the proposed algorithm is stable.In the experiments,we compare the proposed model with other three model.And the proposed model is better than the comparison models in terms of the indicators and the visually.This illustrates the effectiveness of the proposed model in this thesis.