Abstract:Tensor completion estimates missing components by exploiting the low-rank structure of multi-way data. The recently proposed methods based on tensor train (TT) and tensor ring (TR) show better performance in image recovery than classical ones. Compared with TT and TR, the projected entangled pair state (PEPS), which is also called tensor grid (TG), allows more interactions between different dimensions, and may lead to more compact representation. In this paper, we propose to perform image completion based on low-rank tensor grid. A two-stage density matrix renormalization group algorithm is used for initialization of TG decomposition, which consists of multiple TT decompositions. The latent TG factors can be alternatively obtained by solving alternating least squares problems. To further improve the computational efficiency, a multi-linear matrix factorization for low rank TG completion is developed by using parallel matrix factorization. Experimental results on synthetic data and real-world images show the proposed methods outperform the existing ones in terms of recovery accuracy.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the low - rank tensor representation problem in image completion. Specifically, the author proposes an image completion method based on low - rank tensor grid (TG) to improve the recovery accuracy of missing data. ### Problem Background In the field of image processing, tensor completion estimates the missing components by using multi - dimensional data structures. Traditional tensor decomposition methods such as CP decomposition, Tucker decomposition and t - SVD perform well when dealing with high - order tensors, but in recent years, tensor - network - based methods such as tensor train (TT) and tensor ring (TR) decomposition have shown better performance. However, these methods still have limitations in the interaction between different dimensions. ### Research Motivation Compared with TT and TR, the projected entangled pair state (PEPS), also known as tensor grid (TG), allows more interactions between dimensions and may provide a more compact representation. Therefore, the author proposes to use low - rank tensor grid for image completion, aiming to further improve the recovery accuracy. ### Main Contributions 1. **Propose an image completion method based on low - rank tensor grid**: The author introduces the tensor grid (TG) decomposition and develops two algorithms - the alternating least squares method (ALS - TG) and the parallel matrix decomposition method (PMac - TG) to optimize the TG factors. 2. **Initialization method**: To improve the computational efficiency, the author proposes a two - stage density matrix renormalization group algorithm (2SDMRG) for the initialization of TG decomposition. 3. **Experimental verification**: Through experiments on synthetic data and real - world images, it is proved that the proposed method is superior to the existing methods in recovery accuracy. ### Mathematical Formulas - The Frobenius norm of a tensor is defined as: \[ \|X\|_F=\sqrt{\langle X, X \rangle}=\sqrt{\sum_{i_1 = 1}^{I_1}\cdots\sum_{i_D = 1}^{I_D}x_{i_1\cdots i_D}^2} \] - The Hadamard product (element - wise product) is defined as: \[ (X\circledast Y)_{i_1\cdots i_D}=x_{i_1\cdots i_D}\cdot y_{i_1\cdots i_D} \] ### Conclusion Through the above methods, the author shows the superior performance of low - rank tensor grid in the image completion task, especially significantly superior to the existing tensor decomposition methods in terms of recovery accuracy.

Low-rank Tensor Grid for Image Completion

High-dimension Tensor Completion via Gradient-based Optimization Under Tensor-train Format

Tensor completion via joint reweighted tensor Q -nuclear norm for visual data recovery

Hierarchical Tensor Ring Completion

Higher-dimension Tensor Completion via Low-rank Tensor Ring Decomposition

Trainable Subspaces for Low Rank Tensor Completion: Model and Analysis

Low-Rank Tensor Completion Using Matrix Factorization Based on Tensor Train Rank and Total Variation

Tensor Factorization for Low-Rank Tensor Completion.

Imbalanced low-rank tensor completion via latent matrix factorization

Tensor Completion Via Collaborative Sparse and Low-Rank Transforms

Multi-Dimensional Visual Data Completion via Low-Rank Tensor Representation Under Coupled Transform

Tensor Completion via a Low-Rank Approximation Pursuit

Visual Data Completion Via Local Sensitive Low Rank Tensor Learning.

Low tensor-ring rank completion: parallel matrix factorization with smoothness on latent space

Two New Low Rank Tensor Completion Methods Based on Sum Nuclear Norm

Robust Low-Rank Tensor Completion Based on Tensor Ring Rank via $\ell _{p,\epsilon }$-Norm

Low rank tensor completion with sparse regularization in a transformed domain

T-product factorization based method for matrix and tensor completion problems

Low-rank tensor completion via tensor joint rank with logarithmic composite norm

Low-Rank Tensor Completion by Truncated Nuclear Norm Regularization

A Novel Tensor Ring Sparsity Measurement for Image Completion