Fast and Accurate Tensor Completion with Total Variation Regularized Tensor Trains

Fast and Accurate Tensor Completion with Total Variation Regularized Tensor Trains

Ko, Ching Yun (The University of Hong Kong)
Batselier, K. (TU Delft Team Jan-Willem van Wingerden)
Daniel, Luca (Massachusetts Institute of Technology)
Yu, Wenjian (Tsinghua University)
Wong, Ngai (The University of Hong Kong)

2020

We propose a new tensor completion method based on tensor trains. The to-be-completed tensor is modeled as a low-rank tensor train, where we use the known tensor entries and their coordinates to update the tensor train. A novel tensor train initialization procedure is proposed specifically for image and video completion, which is demonstrated to ensure fast convergence of the completion algorithm. The tensor train framework is also shown to easily accommodate Total Variation and Tikhonov regularization due to their low-rank tensor train representations. Image and video inpainting experiments verify the superiority of the proposed scheme in terms of both speed and scalability, where a speedup of up to 155\times is observed compared to state-of-the-art tensor completion methods at a similar accuracy. Moreover, we demonstrate the proposed scheme is especially advantageous over existing algorithms when only tiny portions (say, 1%) of the to-be-completed images/videos are known.

image restoration
Tensor completion
tensor-train decomposition
total variation

http://resolver.tudelft.nl/uuid:20dba0ef-6568-44e1-9f62-4a9fd7ed4d70

https://doi.org/10.1109/TIP.2020.2995061

2020-12-21

1057-7149

IEEE Transactions on Image Processing, 29, 6918-6931

Green Open Access added to TU Delft Institutional Repository ‘You share, we take care!’ – Taverne project https://www.openaccess.nl/en/you-share-we-take-care Otherwise as indicated in the copyright section: the publisher is the copyright holder of this work and the author uses the Dutch legislation to make this work public.

Institutional Repository

journal article

© 2020 Ching Yun Ko, K. Batselier, Luca Daniel, Wenjian Yu, Ngai Wong