This package is an attempt to process images taken with a wide-angle lens and deduce the approximate angle of distortion using a genetic algorithm. Once the angle of distortion is deduced then the image can be corrected using a conformal mapping.

At a high level the approach works as follows:

1. From initial image, perform a Hough Transform to detect all straight lines
2. Create an initial set of line pairs by randomly shuffling all detected lines and pairing neighbors together
3. Compute the slope of each line, and then compute the angle between slopes of each pair (loss function)
4. Create parent sets of line pairs determined by the pairs with the smallest angles
5. From the set of parents, create children pairs with random mutations of the start/stop points
6. Compute the new loss function
7. Recurse on steps 4 - 6

The interpretation of the entire algorithm is that once recursion has terminated then all of the pairs of lines are either parallel or quasi-parallel.

If the original image had a wide-angle distortion then the final pairs of lines will not be exactly parallel, and will instead have a slight angular separation between them. This average angular separation is interpreted as the angle of distortion from the photo lens. Once this angle is determined then the original image can have the distortion removed by some conformal mapping (TBD).

Note: This is intended as an experiment and learning experience with genetic algorithms and is not intended as a definite solution to determining arbitrary wide-angle lens distortions. If you would like to contribute to this project or offer insight please open a ticket or contact me directly. Any contributions are greatly appreciated.

Checkout the package and install dependencies by running

git clone https://github.com/brainsqueeze/Image_correction.git
cd ${HOME}/Image_correction
pip install -r requirements.txt

The learning algorithm can be run on the provided sample image by running

python -m src.find_parallel

1. The final conformal mapping on the original image is not yet implemented.
2. There are improvements needed for pruning poorly performing species during each generational epoch.