Topological Invariant-Based Iris Identification via Digital Homology and Machine Learning
To address model redundancy, poor interpretability, and high computational demands of deep learning methods in iris recognition, this paper introduces formal digital homology theory—novel in biometric recognition—and proposes a lightweight iris representation method based on topological invariants (Betti numbers). Standardized iris images are partitioned into grids to extract robust topological features, which are subsequently reduced via PCA and classified using conventional classifiers such as logistic regression. Experiments demonstrate that logistic regression achieves 97.78±0.82% accuracy—significantly outperforming CNNs under identical configuration—while exhibiting lower variance and enabling efficient inference on CPU-only hardware. The approach delivers high accuracy, strong interpretability, minimal computational overhead, and effective small-sample adaptation. It establishes a new paradigm for resource-constrained or high-security authentication scenarios.