This project concerns the creation of a complete biometric system, based entirely on the use of iris images (in visible wavelength).
The project is mainly structured in two modules:
- Segmentation: the "ISis v.2" method and the "Hough approach" method are proposed (code visible in "
SegmentationModule
" directory). - Feature extraction: several models based on CNNs are proposed, finetuned and trained using a few-shot learning approach, through siamese networks (code visible in "
Notebook
" directory).
This project also includes the implementation of a "demo" of the proposed system based on a subset of the proposed models, namely "Hough approach" for segmentation and "featNet" trained from scratch for feature extraction (code visible in "Demo
" directory).
In each directory there is a more specific README.md
file, containing detailed information on how to use that specific code (e.g. in Segmentation
directory is explained how to compile the code and how it's structured)