Effective Tensor Completion via Element-wise Weighted Low-rank Tensor Train with Overlapping Ket Augmentation
Yang Zhang, Yao Wang, Zhi Han, Xi’ai Chen, Yandong Tang
MATLAB2021a with
- Parallel Computing Toolbox
- Image Processing Toolbox
- Statistic and Machine Learning Toolbox
The code was tested on Windows 10 with Intel(R) Core(TM) i7-9700 CPU @ 3.00GHz.
- Run TestOKA.m for checking the Overlapping Ket Augmentation procedure.
- Run Demo_256x256x3_Lena_TMacTT_OKA.m for Lena of size [256, 256, 3] with 90% elements missing.
- Run Demo_256x256x3_Peppers_TMacTT_OKA.m for Peppers of size [256, 256, 3] with 90% elements missing.
- Run Demo_256x256x3_Lena_TWMacTT_KA.m for Lena of size [256, 256, 3] with 90% elements missing.
- Run Demo_256x256x3_Peppers_TWMacTT_KA.m for Peppers of size [256, 256, 3] with 90% elements missing.
- Run Demo_91x111x3_Lena.m for a randomly cropped image Lena of size [91, 111, 3].
- Run Demo_256x256x3_Lena_TWMacTT_OKA.m for Lena of size [256, 256, 3] with 90% elements missing.
- Run Demo_256x256x3_Peppers_TWMacTT_OKA.m for Peppers of size [256, 256, 3] with 90% elements missing.
For other input images, please tune the hyper-parameter thl
in the Demo script to obtain the best performance.
The completion result of the cropped Lena of size 91×111×3 with missing rate 80% is shown as follows.
Algorithms | RSE | PSNR | SSIM |
---|---|---|---|
FBCP | 0.1345 | 22.8181 | 0.6847 |
SiLRTC | 0.1772 | 20.4194 | 0.6191 |
STDC | 0.2833 | 16.3785 | 0.5414 |
TMac | 0.1941 | 19.6644 | 0.5286 |
TMac-TT+OKA | 0.0738 | 28.0621 | 0.9042 |
TWMac-TT+OKA | 0.0689 | 28.6562 | 0.9103 |