Day and Night-Time Dehazing by Local Airlight Estimation.

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Hazy conditions significantly degrade image quality in both daytime and nighttime scenarios.
• Existing dehazing methods, like dark-channel prior, struggle with non-uniform lighting common in nighttime scenes.

Purpose of the Study:

• To develop an effective fusion-based technique for enhancing both day-time and night-time hazy scenes.
• To improve upon existing methods by addressing the challenge of non-uniform airlight estimation in hazy images.

Main Methods:

• Introduced a fusion-based approach that estimates airlight on image patches rather than the entire image.
• Employed multiple patch sizes for local airlight estimation to balance spatial adaptation and accuracy.
• Integrated the discrete Laplacian of the original image and normalized weight maps into a multi-scale fusion process using Laplacian pyramid decomposition.

Main Results:

• The proposed method effectively enhances both day-time and night-time hazy scenes.
• Demonstrated superior performance compared to recent techniques in terms of output quality.
• Achieved significant computational efficiency.

Conclusions:

• The local airlight estimation and multi-scale fusion strategy is effective for image dehazing.
• The technique offers a robust solution for improving visibility in challenging atmospheric conditions.
• The method provides a computationally efficient and high-quality approach to image enhancement.