Rank-1 tensor decomposition for hyperspectral image denoising with nonlocal low-rank regularization

Jize Xue, Yongqiang Zhao

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

15 Scopus citations

Abstract

In hyperspectral imagery denoising, rank-1 tensor decomposition (R1TD) model can utilize the spatial and spectral information jointly and reduce the noise efficiently. It is difficult to estimate the rank of hyperspectral imagery accurately, and the rank uncertainty will make the R1TD denoising algorithm inefficient. The nonlocal similar patches have lower rank than image, it can be used in rank-1 tensor decomposition process instead of explicitly estimating rank parameters. In this work, a nonlocal low-rank regularization is introduced to avoid the rank uncertainty to influence denoising performance. Then an alternating direction method of multipliers (ADMM) optimization technique is designed to solve the minimum problem. Compared with the state of art methods, proposed algorithm significantly improves the hyperspectral imagery quality both in visual inspection and image quality indices.

Original languageEnglish
Title of host publicationProceedings - 2017 International Conference on Machine Vision and Information Technology, CMVIT 2017
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages40-45
Number of pages6
ISBN (Electronic)9781509049936
DOIs
StatePublished - 14 Mar 2017
Event2017 International Conference on Machine Vision and Information Technology, CMVIT 2017 - Singapore, Singapore
Duration: 17 Feb 201719 Feb 2017

Publication series

NameProceedings - 2017 International Conference on Machine Vision and Information Technology, CMVIT 2017

Conference

Conference2017 International Conference on Machine Vision and Information Technology, CMVIT 2017
Country/TerritorySingapore
CitySingapore
Period17/02/1719/02/17

Keywords

  • hyperspectral image denoising
  • low-rank regularization
  • nonlocal patches grouping (NPG)
  • rank estimation bias
  • rank-1 tensor decomposition (R1TD)

Fingerprint

Dive into the research topics of 'Rank-1 tensor decomposition for hyperspectral image denoising with nonlocal low-rank regularization'. Together they form a unique fingerprint.

Cite this