This project explores the comparison between traditional and deep learning-based approaches for iris recognition.
The classical pipeline, implemented using the OpenIris library, relies on Gabor filters for feature extraction and Hamming distance for matching.
In contrast, the deep learning model leverages a ResNet-50 architecture, generating feature embeddings and evaluating similarity using cosine metrics.
Both methods were evaluated on the CASIA-Iris-Thousand dataset under identical conditions, highlighting the trade-offs between rule-based and data-driven techniques for biometric authentication.