Joint Texture Search and Histogram Redistribution for Hyperspectral Image Quality Improvement

Area of Science:

• Remote Sensing
• Image Processing
• Data Science

Background:

• Hyperspectral remote sensing data is susceptible to various noises (optical, electrical, compression).
• Noise contamination significantly impacts the utility of hyperspectral data in applications.
• Band-wise processing methods are inadequate for hyperspectral data due to spectral accuracy concerns.

Purpose of the Study:

• To develop a quality enhancement algorithm for hyperspectral imaging data.
• To address noise reduction and contrast enhancement while preserving spectral information.
• To improve the overall usability of hyperspectral datasets for analysis.

Main Methods:

• A novel texture-based search algorithm is proposed to enhance denoising accuracy via improved sparsity in 4D block matching clustering.
• Histogram redistribution and Poisson fusion techniques are employed for spatial contrast enhancement.
• The algorithm combines denoising and contrast enhancement without compromising spectral fidelity.

Main Results:

• Quantitative evaluation using synthesized noisy hyperspectral data demonstrates satisfactory performance.
• Multiple criteria analysis confirms the algorithm's effectiveness in improving data quality.
• Classification tasks utilizing the enhanced data show improved performance, validating the data quality enhancement.

Conclusions:

• The proposed algorithm effectively enhances hyperspectral data quality by mitigating noise and improving contrast.
• The method successfully preserves spectral accuracy, making it suitable for advanced hyperspectral data processing.
• This approach offers a significant improvement for applications relying on high-quality hyperspectral imaging data.