Nonlocal Sparse Tensor Factorization for Semiblind Hyperspectral and Multispectral Image Fusion.

Area of Science:

• Remote Sensing
• Image Processing
• Data Fusion

Background:

• Hyperspectral image (HSI) spatial resolution enhancement commonly uses high-spatial-resolution multispectral images (HR-MSI).
• Existing fusion methods based on matrix factorization may not fully exploit the 3D structure of HSI data.
• Nonlocal self-similarities in HSI are a key characteristic for advanced fusion techniques.

Purpose of the Study:

• To develop a novel semiblind fusion method for HSI and MSI.
• To improve upon existing matrix factorization techniques by utilizing tensor decomposition.
• To address the limitations of current methods in handling sensor point spread functions (PSFs).

Main Methods:

• Proposed a nonlocal sparse tensor factorization approach (NLSTF_SMBF) for HSI and MSI fusion.
• Decomposed HSI into full-band patches (FBPs) and factored them into dictionaries and a sparse core tensor.
• Clustered similar FBPs to share dictionaries, exploiting nonlocal self-similarities and performing tensor sparse coding.

Main Results:

• The NLSTF_SMBF method demonstrated superior performance compared to state-of-the-art fusion techniques.
• The method effectively handles semiblind fusion scenarios, including spatially variant point spread functions (PSFs).
• Experimental results validated the advantages of the proposed tensor factorization approach.

Conclusions:

• NLSTF_SMBF offers a robust and effective solution for enhancing HSI spatial resolution.
• The method's ability to be blind to the sensor's PSF is a significant advancement.
• This approach provides a powerful tool for analyzing hyperspectral and multispectral data in complex imaging conditions.