Abstract:This paper addresses the tensor completion problem, which aims to recover missing information of multi-dimensional images. How to represent a low-rank structure embedded in the underlying data is the key issue in tensor completion. In this work, we suggest a novel low-rank tensor representation based on coupled transform, which fully exploits the spatial multi-scale nature and redundancy in spatial and spectral/temporal dimensions, leading to a better low tensor multi-rank approximation. More precisely, this representation is achieved by using two-dimensional framelet transform for the two spatial dimensions, one/two-dimensional Fourier transform for the temporal/spectral dimension, and then Karhunen–Loéve transform (via singular value decomposition) for the transformed tensor. Based on this low-rank tensor representation, we formulate a novel low-rank tensor completion model for recovering missing information in multi-dimensional visual data, which leads to a convex optimization problem. To tackle the proposed model, we develop the alternating directional method of multipliers (ADMM) algorithm tailored for the structured optimization problem. Numerical examples on color images, multispectral images, and videos illustrate that the proposed method outperforms many state-of-the-art methods in qualitative and quantitative aspects.

What problem does this paper attempt to address?

This paper aims to solve the tensor completion problem in multi - dimensional images. The specific goal is to recover the missing information in multi - dimensional visual data from a small number of known observations. The key to the paper lies in how to represent the low - rank structure embedded in the underlying data, which is the core problem of tensor completion. The author proposes a new low - rank tensor representation method based on coupled transformation. This method makes full use of the spatial multi - scale characteristics and the redundancy of spatial and spectral / temporal dimensions, thereby achieving a better multi - rank approximation of low - rank tensors. ### Main contributions: 1. **New low - rank tensor representation**: A low - rank tensor representation method based on coupled transformation is proposed, which can more effectively capture the internal structure in multi - dimensional visual data, especially achieving low - rank approximation in the transform domain. 2. **Low - rank tensor completion model**: Based on the proposed low - rank tensor representation method, a new low - rank tensor completion model is constructed, and an efficient alternating direction multiplier method (ADMM) algorithm is developed to solve this model. 3. **Experimental verification**: Through numerical experiments on color images, multispectral images and videos, it is proved that the proposed method is superior to many existing methods in both qualitative and quantitative aspects. ### Method overview: 1. **First - layer transformation (B1)**: The two - dimensional framelet transform is used to describe the local spatial correlation, which can express the rich spatial structures at different frequencies and spatial directions in detail, and the multi - level decomposition helps to capture the multi - scale features of the tensor. 2. **Second - layer transformation (B2)**: The Fourier transform is used to depict the global correlation in the time (or spectral) dimension, revealing the strong correlation between the multi - scale representation coefficients after the first - layer transformation. 3. **Third - layer transformation (B3)**: The Karhunen - Loève transform (KL transform) is used to characterize the global spatial correlation of the results of the first two - layer transformations through singular value decomposition (SVD), combining global correlation and local geometric details. ### Mathematical model: The proposed low - rank tensor completion model can be expressed as: \[ \min_{X} \| B(X) \|_1 \quad \text{s.t.} \quad P_\Omega(X) = P_\Omega(O) \] where \( B = B_3 \circ B_2 \circ B_1 \) is the coupled transformation, involving three - layer transformations, namely framelet transform, Fourier transform and KL transform. \( P_\Omega \) is the sampling operator, which retains the elements in the index set \( \Omega \) and sets other positions to zero. ### Experimental results: The experimental results show that the proposed method performs well on a variety of multi - dimensional visual data, especially at a low sampling rate, and can better recover specific textures rather than just the global outline. ### Conclusion: This paper effectively solves the tensor completion problem in multi - dimensional visual data by proposing a low - rank tensor representation method based on coupled transformation, which has high practical value and theoretical significance.

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

Visual Data Completion Via Local Sensitive Low Rank Tensor Learning.

Thick Cloud Removal in Multi-Temporal Remote Sensing Images Via Frequency Spectrum-Modulated Tensor Completion

Low Multilinear Rank Tensor Completion with Missing Data

Trainable Subspaces for Low Rank Tensor Completion: Model and Analysis

Low rank tensor completion with sparse regularization in a transformed domain

Multi-Tensor Network Representation for High-Order Tensor Completion

Tensor Completion Via Collaborative Sparse and Low-Rank Transforms

Beyond Low Rank: A Data-Adaptive Tensor Completion Method

CoNoT: Coupled Nonlinear Transform-Based Low-Rank Tensor Representation for Multidimensional Image Completion

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

Tensor Completion Via a Multi-Linear Low-N-rank Factorization Model

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

Tensor Completion via a Low-Rank Approximation Pursuit

Tensor Completion via Complementary Global, Local, and Nonlocal Priors

Tensor completion via multi-directional partial tensor nuclear norm with total variation regularization

Tensor completion using total variation and low-rank matrix factorization

Auto-weighted robust low-rank tensor completion via tensor-train

A Biased Deep Tensor Factorization Network For Tensor Completion

Accurate Tensor Completion via Adaptive Low-Rank Representation

Non-local tensor completion for multitemporal remotely sensed images inpainting