Regularization Parameter Estimation for Non-Negative Hyperspectral Image Deconvolution

Area of Science:

• Image processing
• Computational imaging
• Optimization techniques

Background:

• Hyperspectral image deconvolution is crucial for analyzing spectral data.
• Estimating regularization parameters is a key challenge in non-negative deconvolution.
• Existing methods may lack efficiency or theoretical guarantees.

Purpose of the Study:

• To develop and evaluate automatic methods for estimating regularization parameters in non-negative hyperspectral image deconvolution.
• To introduce the Minimum Distance Criterion (MDC) and Maximum Curvature Criterion (MCC) for this purpose.
• To compare these new criteria against state-of-the-art methods.

Main Methods:

• Formulating deconvolution as a multi-objective optimization problem.
• Analyzing the properties of the response surface.
• Proposing MDC and MCC based on these properties.
• Implementing a grid-search approach for computational efficiency.

Main Results:

• Both MDC and MCC effectively estimate regularization parameters for non-negativity constrained deconvolution.
• Fast MDC and fast MCC approaches significantly reduce computational cost.
• Simulated 2D image analysis shows MDC and MCC outperform existing methods.
• For non-negative hyperspectral deconvolution, fast MDC shows better performance than fast MCC.

Conclusions:

• The proposed MDC and MCC offer fast and effective solutions for regularization parameter estimation in non-negative hyperspectral image deconvolution.
• MDC, particularly the fast version, is recommended for its superior performance and theoretical advantages.
• The methods are validated on both simulated and real-world hyperspectral fluorescence microscopy data.