Haze removal with channel-wise scattering coefficient awareness based on grey pixels

Area of Science:

• Computer Vision
• Image Processing
• Atmospheric Optics

Background:

• Light information is attenuated by transmission media, a process dependent on wavelength and distance.
• Existing haze removal techniques often overlook wavelength dependency, leading to color distortions.

Purpose of the Study:

• To develop an advanced haze removal method addressing wavelength-dependent atmospheric scattering.
• To improve image quality by correcting color distortions caused by haze.

Main Methods:

• A scattering coefficient aware method based on the image formation model.
• Initial transmission estimation using dark channel prior.
• Calculation of scattering coefficient ratios and fine transmission maps for RGB channels.
• Compensation for sky areas and global correction for color bias.

Main Results:

• The proposed method effectively handles wavelength-dependent attenuation.
• Improved restoration of images with white fog and colorized haze.
• Quantitative and qualitative comparisons show superior performance against state-of-the-art methods.

Conclusions:

• The scattering coefficient aware method offers superior haze removal, particularly for color-distorted images.
• This approach enhances image quality in diverse atmospheric conditions.
• It provides a significant advancement in addressing limitations of current haze removal techniques.