Noisy Tensor Completion via Low-Rank Tensor Ring

This repository implements a novel noisy tensor completion model for recovering incomplete high-order tensor data with noise.

Overview

NTRC (Noisy Tensor Ring Completion) is a tensor completion algorithm that combines tensor ring nuclear norm with least-squares estimation to effectively handle noisy and incomplete tensor data. FaNTRC (Fast NTRC) accelerates the original NTRC method by equivalently minimizing the tensor ring nuclear norm on a smaller core tensor through heterogeneous tensor decomposition, making it particularly efficient for large-scale tensor completion tasks.

Requirements

• MATLAB R2019b or later
• Tensor Toolbox 3.6 (included in tensor_toolbox3.6 directory)

Key parameters:

• sr: Sampling rate
• c: Noise level
• lambda: Regularization parameter
• R0: Tensor ranks for Faster NTRC

Data Preparation

• Video data should be placed in data/YUV/
• Color images should be placed in data/Images512/
• Light field images should be placed in data/LightField/

Citation

If you use this code in your research, please cite:

@ARTICLE{9800181,
author={Qiu, Yuning and Zhou, Guoxu and Zhao, Qibin and Xie, Shengli},
journal={IEEE Transactions on Neural Networks and Learning Systems},
title={Noisy Tensor Completion via Low-Rank Tensor Ring},
year={2024},
volume={35},
number={1},
pages={1127-1141},
doi={10.1109/TNNLS.2022.3181378}
}

License

This project is licensed under the MIT License - see the LICENSE file for details.