Hyperspectral Image Fusion via a Novel Generalized Tensor Nuclear Norm Regularization.

Area of Science:

• Remote Sensing
• Image Processing
• Data Fusion

Background:

• Low-rank tensor regularization is increasingly used in hyperspectral and multispectral fusion (HMF).
• Existing methods face challenges with inflexible tensor definitions and sensitivity to mode permutations.
• These limitations hinder optimal performance in HMF tasks.

Purpose of the Study:

• To propose a novel generalized tensor nuclear norm (GTNN) for improved HMF.
• To address the inflexibility and permutation sensitivity of current low-rank tensor methods.
• To enhance the accuracy and robustness of hyperspectral and multispectral image fusion.

Main Methods:

• A novel generalized tensor nuclear norm (GTNN) is defined by extending the tensor nuclear norm (TNN) to arbitrary modes via Fourier transform.
• High-resolution hyperspectral images (HSI) are modeled using low-rank spectral basis multiplication, with spectral basis estimated via SVD.
• GTNN regularization is applied to clustered coefficient patches, exploiting non-local spatial-spectral similarities for robust fusion.

Main Results:

• The proposed GTNN effectively captures extensive correlations across tensor modes.
• The method demonstrates reduced sensitivity to tensor mode permutations, enhancing stability.
• Fusion experiments on simulated and real datasets validate the superior performance of the GTNN-based approach.

Conclusions:

• The novel GTNN provides a more flexible and robust regularization technique for HMF.
• This approach successfully models non-local self-similarities within HSI data.
• The developed method offers significant advantages for hyperspectral and multispectral fusion applications.