Haze Airlight Color Estimation

Matlab code implementing the Airlight color estimation part of the ICCP 2016 paper "Blind Dehazing Using Internal Patch Recurrence". Tested on Matlab 2016a working on Windows.

If you find our work useful in your research or publication, please cite it:

@inproceedings{bahat2016blind,
  title={Blind dehazing using internal patch recurrence},
  author={Bahat, Yuval and Irani, Michal},
  booktitle={Computational Photography (ICCP), 2016 IEEE International Conference on},
  pages={1--9},
  year={2016},
  organization={IEEE}
}

Functions:

1. AirlightUsingPatchRecurrence(orgImage):

The core function estimating the haze airlight color in the input image. The input image should be in the form of 3 (RGB) channels (double, values in the range of [0,1]).

2. testAirlightEstimation:

A script for testing the airlight estimation. Estimated airlight is presented alongside the corresponding image. This script also displays a ground truth airlight color, either saved in the GTairlights folder or manually extracted by the user.

3. estimate4allImages:

Serially computing the airlight color for all images in the 'images' sub-folder. Comparing the estimated airlights with the pre-determined GT airlights for images that contain a region of sky, and presenting the mean distance in RGB space.

Further comments:

The 'images' sub-folder contain example images, and the 'GTairlights' sub-folder contains ground truth airlight colors for images containing regions of sky. This code uses the nearest neighbors search engine by Shai Bagon, named ANN. For convinience, it appears as-is in the ann_wrapper sub-folder.

The code is provided as-is for academic use only and without any guarantees. Please contact the author to report any bugs. Written by Yuval Bahat.