Zero-Shot Image Dehazing.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Image Processing

Background:

• Single image dehazing is crucial for improving visual quality.
• Existing methods often require paired hazy and clean images or large datasets.
• Unsupervised and zero-shot approaches are needed to overcome data limitations.

Purpose of the Study:

• To develop an unsupervised and zero-shot method for single image dehazing.
• To address the challenges of data collection and domain shift in image dehazing.
• To propose a novel layer disentanglement approach for haze removal.

Main Methods:

• Proposed ZID (Zero-shot Image Dehazing) method.
• Viewed hazy images as entangled layers: hazy-free image, transmission map, and atmospheric light.
• Employed layer disentanglement for unsupervised and zero-shot learning.

Main Results:

• Achieved promising performance in qualitative and quantitative evaluations.
• Outperformed 15 existing single image dehazing approaches.
• Demonstrated effectiveness without relying on ground-truth clean images or large datasets.

Conclusions:

• ZID offers an effective unsupervised and zero-shot solution for single image dehazing.
• The layer disentanglement approach avoids data collection and domain shift issues.
• The method shows significant potential for real-world image restoration applications.