Compressive hyperspectral imaging with non-zero mean noise

Area of Science:

• Optics
• Image Processing
• Spectroscopy

Background:

• Compressive hyperspectral imaging (CHI) offers snapshot acquisition by encoding 3D spectral data into 2D measurements.
• Current CHI reconstruction quality is hindered by inaccurate noise modeling, particularly the assumption of zero-mean noise.

Purpose of the Study:

• To address the limitations of traditional zero-mean noise models in CHI.
• To develop a novel reconstruction method that accounts for non-zero noise mean in practical imaging conditions.

Main Methods:

• Proposed a generalized noise model incorporating a non-zero mean to represent real-world imaging artifacts like stray light and dark current.
• Introduced an iterative reconstruction algorithm that estimates and refines the noise mean concurrently with the spectral image reconstruction.
• Validated the method using both synthetic datasets and experimental data acquired with a hardware prototype.

Main Results:

• The proposed non-zero mean noise model more accurately reflects optical imaging principles.
• The novel reconstruction method demonstrated superior performance compared to traditional approaches.
• Significant improvements in hyperspectral image reconstruction quality were observed in experimental results.

Conclusions:

• Accurate noise modeling, specifically accounting for non-zero mean, is crucial for enhancing CHI reconstruction quality.
• The developed method offers a practical solution for improving the fidelity of hyperspectral imaging systems.
• This work paves the way for more reliable applications of CHI in various scientific fields.