A Bayesian framework for single image dehazing considering noise

Area of Science:

• Computer Vision
• Image Processing
• Computational Photography

Background:

• Existing single image dehazing algorithms prioritize haze removal efficiency over noise reduction.
• Image degradation often involves both atmospheric haze and sensor noise, complicating restoration.
• A unified approach is needed to address both dehazing and denoising concurrently.

Purpose of the Study:

• To propose a Bayesian framework for single image dehazing that incorporates noise considerations.
• To develop an algorithm capable of simultaneously removing haze and noise from degraded images.
• To achieve a balance between dehazing performance and noise suppression.

Main Methods:

• A Bayesian framework was adapted for image dehazing.
• The probability density function of an improved atmospheric scattering model was estimated using statistical priors and image assumptions.
• An iterative feedback approach was employed to refine the reflectance image.

Main Results:

• The proposed method effectively removes haze from single images.
• Simultaneous removal of noise alongside haze was achieved.
• Experimental results validated the method's ability to balance dehazing and denoising.

Conclusions:

• The developed Bayesian framework offers a robust solution for single image dehazing with integrated denoising.
• The iterative approach ensures effective simultaneous removal of haze and noise.
• This method enhances image quality by addressing multiple degradation factors.