yzcv / TWMac-TT-OKA

Official implementation of TWMac-TT-OKA.

TWMac-TT-OKA

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

[Paper Link]

1. Environment Requirements

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.

2. Test

OKA quick test

• Run TestOKA.m for checking the Overlapping Ket Augmentation procedure.

TMacTT-OKA (Only OKA)

• 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.

TWMacTT-KA (Only W)

• 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.

TWMacTT-OKA (W+OKA) - arbitrary size test (thanks to the proposed OKA scheme

• Run Demo_91x111x3_Lena.m for a randomly cropped image Lena of size [91, 111, 3].

TWMacTT-OKA (W+OKA)

• 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.

3. Representative Result

The completion result of the cropped Lena of size 91×111×3 with missing rate 80% is shown as follows.

Visual performance

Numeric metrics

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