Nonconvex 3D array image data recovery and pattern recognition under tensor framework

作者：

Highlights：

• A new weighted tensor Schatten-p quasi-norm regularizer is proposed to approximate the rank of multi-dimensional data sets.

• The proposed framework is shown to be suitable for solving both tensor completion and tensor robust principal component analysis problems.

• An efficient algorithm based on the alternation direction method of multipliers is established and detailed mathematical justification is provided.

• Extensive evaluation on several benchmark data sets along with the comparison with latest development is presented.

摘要

•A new weighted tensor Schatten-p quasi-norm regularizer is proposed to approximate the rank of multi-dimensional data sets.•The proposed framework is shown to be suitable for solving both tensor completion and tensor robust principal component analysis problems.•An efficient algorithm based on the alternation direction method of multipliers is established and detailed mathematical justification is provided.•Extensive evaluation on several benchmark data sets along with the comparison with latest development is presented.

论文关键词：Tensor completion,Tensor robust principle component analysis,T-SVD,Tensor nuclear norm,Weighted tensor schatten-p quasi-norm

论文评审过程：Received 16 December 2020, Revised 28 August 2021, Accepted 9 September 2021, Available online 17 October 2021, Version of Record 17 October 2021.

论文官网地址：https://doi.org/10.1016/j.patcog.2021.108311