The aim of this project is to re-create (in the numpiest way) an iris segmentation algorithm. Iris detection is done using Daugman's integro-differential operator [1] on the multi-channel as done by Haindl-Krupička in [2]. Daugman's operator is then used on the red channel to detect the pupil. After doing this, normalization is done with the rubber sheet model and occlusions are detected using methods described in [2].

If it is the first time using weights & biases, you need to set it up.

1. Get a w&b account
2. Install the wandb command line client with pip install wandb inside the virtualenv (i.e. you already ran one of conda activate my_virtual_env_name or source /path/to/my/venv/bin/activate depending on how you set it up)
3. Use the command line utility to log in to your account: wandb login
4. Done! you can run a hyperparameter search with bash run_sweep, and all your work will be logged to you weights & biases account

1. J. G. Daugman, "High confidence visual recognition of persons by a test of statistical independence," in IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 15, no. 11, pp. 1148-1161, Nov. 1993, doi: 10.1109/34.244676
2. Michal Haindl, Mikuláš Krupička, Unsupervised detection of non-iris occlusions, Pattern Recognition Letters, Volume 57, 2015, Pages 60-65, ISSN 0167-8655, https://doi.org/10.1016/j.patrec.2015.02.012