Band Selection for Dehazing Algorithms Applied to Hyperspectral Images in the Visible Range

Area of Science:

• Computer Vision
• Image Processing
• Remote Sensing

Background:

• Degraded image quality (low contrast, visibility, color distortion) in adverse weather conditions (fog, haze, dust) is a significant challenge.
• Image degradation severity is influenced by atmospheric particle density, distance, and wavelength.

Purpose of the Study:

• To evaluate eight single image dehazing algorithms on hazy spectral images.
• To identify optimal wavelength triplets for dehazing using a novel combined image quality metric.
• To assess algorithm performance based on spectral band selection.

Main Methods:

• Analysis of eight single image dehazing algorithms on a hazy spectral image database.
• Brute-force search for optimal three-wavelength combinations.
• Evaluation using a new combined image quality metric.
• Comparison of full-spectrum (450-720 nm) vs. selected bands.

Main Results:

• Optimal wavelength triplets are algorithm-dependent and often spectrally close.
• Artificial Multiple Exposure Image Fusion (AMEF) performed best according to the combined metric.
• AMEF and Contrast Limited Adaptive Histogram Equalization (CLAHE) showed similar high performance for sRGB renderization using the full visible spectrum.
• Algorithm performance is sensitive to signal balance and channel information.

Conclusions:

• Using a triplet of optimized spectral bands can significantly reduce capture time and simplify devices for dehazing.
• Band selection must be tailored to the specific dehazing algorithm employed.
• Algorithm performance is critically dependent on the spectral information utilized.