Contrast in Haze Removal: Configurable Contrast Enhancement Model Based on Dark Channel Prior

Area of Science:

• Computer Vision
• Image Processing
• Computational Photography

Background:

• Conventional haze removal methods often compromise image luminance for contrast enhancement.
• This trade-off can lead to suboptimal image quality and unnatural-looking results.

Purpose of the Study:

• To reformulate haze removal as a luminance reconstruction problem.
• To develop a method that achieves a favorable balance between luminance and contrast.
• To improve the estimation of atmospheric light for more robust haze removal.

Main Methods:

• A novel energy term was formulated to optimize the trade-off between luminance and contrast.
• Luminance values were derived from statistical analysis of haze-free images.
• A new atmospheric light estimation module based on color constancy was developed.

Main Results:

• The proposed method achieves superior contrast for a given brightness level compared to existing techniques.
• The novel atmospheric light estimation module demonstrates improved performance, especially in noisy conditions.
• The framework processes a 1-megapixel image in just 0.55 seconds.

Conclusions:

• The developed haze-removal framework effectively balances luminance and contrast.
• The method provides a theoretically sound approach to image restoration.
• The framework offers a computationally efficient and robust solution for haze removal.