Researchers have developed a new law that quantises the relationship between the number of features used in biometric identification systems and their equal error rate (EER). This law, derived from statistical principles involving Gaussian features and cosine similarity, predicts how many features are needed to achieve a desired EER based on their independence and temporal persistence. The model accurately reproduces existing simulation data and empirical findings from various biometric modalities, including eye movements, keystroke dynamics, electrocardiography, and face embeddings, offering a practical design rule for optimizing biometric systems. AI
RANK_REASON Academic paper detailing a new theoretical law for biometric systems. [lever_c_demoted from research: ic=1 ai=0.4]
- cosine similarity
- electrocardiography
- face embeddings
- Fisher's z-transform
- Gaussian
- GazeBase
- GazeBaseVR
- keystroke dynamics
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →