Image Defogging Framework Using Segmentation and the Dark Channel Prior.

Area of Science:

• Computer Vision
• Image Processing

Background:

• Foggy images exhibit low contrast, poor visibility, and color distortion, necessitating effective image defogging techniques.
• The Dark Channel Prior (DCP) method excels in defogging but struggles with homogeneous regions like skies, causing color distortion and block effects.

Purpose of the Study:

• To develop an improved image defogging framework that overcomes the limitations of the DCP method, particularly in regions with large homogeneous areas.
• To enhance the visual quality and naturalness of defogged images by addressing color distortion and block effects.

Main Methods:

• A novel framework employing sky and non-sky region segmentation using a binary mask generated by the floodfill algorithm.
• Restoring the foggy sky region using Contrast Limited Adaptive Histogram Equalization (CLAHE) and the non-sky region using a modified Dark Channel Prior (DCP) technique.
• Blending the restored sky and non-sky regions to produce the final defogged image.

Main Results:

• The proposed method demonstrates superior performance compared to state-of-the-art techniques on both synthetic and real-world foggy images.
• Achieved higher entropy values, indicating improved image quality and detail preservation.
• Produced defogged images with more natural visual effects and significantly reduced processing time.

Conclusions:

• The proposed segmentation-based defogging framework effectively addresses the limitations of traditional DCP methods, especially in images with sky regions.
• The combination of CLAHE and modified DCP offers a robust solution for enhancing visibility and color fidelity in foggy conditions.
• This approach provides a computationally efficient and visually superior alternative for image defogging in computer vision applications.