Real-World Nighttime Image Dehazing via Bayesian-Based Fractional-Order Variational Model.

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Nighttime hazy images suffer severe degradation (low visibility, color distortion, reduced contrast).
• Existing dehazing methods are primarily designed for daytime and fail with complex nighttime conditions.

Purpose of the Study:

• To develop a novel framework for real-world nighttime image dehazing.
• To address complex degradations including haze, low-light, noise, and glow.

Main Methods:

• A Bayesian-based variational framework with fractional-order constraints.
• Constructed a simplified physical model for nighttime hazy images.
• Used anisotropic pre-processing, MAP estimation, and alternating direction minimization.

Main Results:

• The proposed method significantly outperforms state-of-the-art dehazing techniques.
• Achieved superior qualitative and quantitative results on real-world datasets.
• Demonstrated effective generalization to other degraded scenes and vision tasks.

Conclusions:

• The novel framework effectively handles nighttime image dehazing challenges.
• Fractional-order constraints promote piecewise smoothness and preserve image details.
• The method offers a robust solution for enhancing vision-based tasks in adverse conditions.