Learning a Patch Quality Comparator for Single Image Dehazing.

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Atmospheric particles scatter light, reducing visibility and contrast in images captured during foggy or hazy conditions.
• Hazy images often appear colorless, necessitating image dehazing techniques for restoration.
• Existing image dehazing methods aim to recover clear, haze-free portrayals from degraded imagery.

Purpose of the Study:

• To propose a novel image dehazing method that effectively restores clarity and color to hazy images.
• To develop a robust patch comparison mechanism for selecting the optimal dehazed image segment.
• To evaluate the proposed method's performance against established state-of-the-art techniques.

Main Methods:

• A Convolutional Neural Network (CNN) based patch quality comparator was developed for evaluating dehazed image patches.
• A binary search algorithm is employed to efficiently select the best dehazed patch by comparing outputs with the original hazy image.
• The method iteratively refines the dehazed image by selecting optimal patches.

Main Results:

• The proposed method demonstrates significant improvements in image visibility and contrast restoration.
• Quantitative and qualitative evaluations indicate good performance across various hazy conditions.
• The method's results are found to be comparable, on average, to current state-of-the-art image dehazing approaches.

Conclusions:

• The developed CNN-based comparator and binary search strategy provide an effective approach to image dehazing.
• The method successfully recovers haze-free images, enhancing visual quality.
• The proposed technique offers a competitive alternative to existing image dehazing solutions.