Abstract:This paper expands on the foundational concept of temporal persistence in
biometric systems, specifically focusing on the domain of eye movement
biometrics facilitated by machine learning. Unlike previous studies that
primarily focused on developing biometric authentication systems, our research
delves into the embeddings learned by these systems, particularly examining
their temporal persistence, reliability, and biometric efficacy in response to
varying input data. Utilizing two publicly available eye-movement datasets, we
employed the state-of-the-art Eye Know You Too machine learning pipeline for
our analysis. We aim to validate whether the machine learning-derived
embeddings in eye movement biometrics mirror the temporal persistence observed
in traditional biometrics. Our methodology involved conducting extensive
experiments to assess how different lengths and qualities of input data
influence the performance of eye movement biometrics more specifically how it
impacts the learned embeddings. We also explored the reliability and
consistency of the embeddings under varying data conditions. Three key metrics
(kendall's coefficient of concordance, intercorrelations, and equal error rate)
were employed to quantitatively evaluate our findings. The results reveal while
data length significantly impacts the stability of the learned embeddings,
however, the intercorrelations among embeddings show minimal effect.
What problem does this paper attempt to address?
### Problems the Paper Aims to Solve
This paper aims to explore the temporal persistence, reliability, and biometric performance of embeddings in machine learning-driven eye movement biometric systems. Specifically, the researchers focus on the following aspects:
1. **Temporal Persistence Verification**:
- Verify whether the eye movement biometric embeddings generated by machine learning have the same temporal persistence as traditional biometric features.
- Traditional biometric features typically exhibit high temporal stability, which is crucial for the performance of biometric systems.
2. **Impact of Input Data Variations**:
- Investigate the impact of input data of different lengths and qualities on the temporal stability and overall reliability of the embeddings.
- Explore the reliability and consistency of embeddings under different data conditions.
3. **Performance Evaluation**:
- Quantitatively evaluate the experimental results using three key metrics (Kendall's coefficient of concordance, correlation, equal error rate).
- Analyze the performance of embeddings under different experimental conditions to provide insights into the adaptability and consistency of machine learning models when handling diverse data.
### Background and Motivation
- **Importance of Biometric Systems**: Biometric systems have become the preferred security solution due to their enhanced reliability and convenience.
- **Significance of Temporal Persistence**: Temporal persistence (i.e., temporal stability and permanence) is an important indicator of biometric system performance.
- **Potential of Eye Movement Biometrics**: Eye movement, as a behavioral biometric modality, has garnered widespread attention due to its unique and consistent patterns, offering high specificity, anti-spoofing, and continuous authentication advantages.
- **Transition from Traditional to Machine Learning**: The shift from traditional biometric systems to machine learning-driven biometric systems raises questions about the applicability of temporal persistence in this new domain.
### Research Methods
- **Datasets**: Two publicly available eye movement datasets (GazeBase and GazeBaseVR) were used, collected by high-end eye trackers and virtual reality headsets with eye-tracking capabilities, respectively.
- **Model Architecture**: The state-of-the-art Eye Know You Too (EKYT) model, based on DenseNet, was employed to generate 128-dimensional embeddings.
- **Experimental Design**:
- **Impact of Sequence Length**: Analyze the impact of different sequence lengths (5 seconds, 10 seconds, 30 seconds, 60 seconds) on the embeddings.
- **Impact of Sampling Rate**: Study the impact of different sampling rates (1000Hz, 500Hz, 333Hz, 250Hz, 100Hz, 50Hz, 25Hz, 10Hz) on the embeddings.
- **Impact of Sample Length**: Analyze the impact of reducing sample length during training (100%, 50%, 33.3%, 25%, 10%, 5%, 2.5%, 1%) on the embeddings.
- **Impact of Signal Quality**: Evaluate the impact of spatial precision and spatial accuracy on the embeddings.
### Results
- **Impact of Sequence Length**: Embeddings generated from longer data sequences (e.g., 60 seconds) exhibit higher temporal persistence and lower equal error rate (EER), indicating better biometric performance.
- **Impact of Sampling Rate**: Embeddings generated from higher sampling rates (e.g., 1000Hz) exhibit better temporal persistence and lower EER, but even at lower sampling rates, embeddings can still maintain a certain level of stability and performance.
- **Impact of Sample Length**: Reducing sample length significantly decreases the temporal persistence and biometric performance of the embeddings, especially in extreme cases (e.g., 1% sample length).
- **Impact of Signal Quality**: Embeddings generated from data with higher spatial precision and spatial accuracy exhibit better temporal persistence and lower EER.
### Conclusion
This paper provides important insights into the application of machine learning in the biometric field by systematically analyzing the temporal persistence, reliability, and biometric performance of machine learning-generated eye movement biometric embeddings. The research results indicate that although variations in input data can affect the performance of embeddings, under appropriate conditions, machine learning-generated embeddings can still maintain good temporal persistence and biometric performance.