[A noise reduction algorithm of hyperspectral imagery using double-regularizing terms total variation]

Area of Science:

• Remote Sensing
• Image Processing
• Signal Processing

Context:

• Hyperspectral imagery (HSI) is susceptible to noise in both spatial and spectral domains.
• Existing denoising methods may not optimally address these distinct noise characteristics.
• Accurate noise reduction is crucial for reliable HSI analysis and interpretation.

Purpose:

• To propose an effective total variation (TV) denoising algorithm tailored for hyperspectral imagery.
• To generalize the classical 2D TV denoising to a 3D formulation for HSI data.
• To improve the objective function by incorporating separate spatial and spectral regularization terms.

Summary:

• A novel 3D total variation denoising algorithm is developed for hyperspectral imagery.
• The algorithm utilizes double-regularizing terms (spatial and spectral) and separate parameters to address distinct noise characteristics.
• A majorization-minimization (MM) based iteration is employed to minimize a convex quadratic function, enabling independent noise removal.
• Experiments on Hyperion data demonstrate improved signal-to-noise ratio and better spectral absorption peak restoration compared to existing methods.

Impact:

• The proposed algorithm offers a significant improvement in hyperspectral image denoising.
• It achieves comparable signal-to-noise ratio enhancement to established methods like Minimum Noise Fraction (MNF) and Savitzky-Golay filter.
• Crucially, it excels at removing spectral indentations and restoring vital spectral absorption features, enhancing data utility.