Fingerprint Model Dynamic Update for Fake Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Fingerprint recognition systems struggle to distinguish between genuine and fake fingerprints, particularly in changing environments, leading to misrecognition issues due to the inability to cover all actual use environments with training databases.
Innovation Solution
A processor-implemented method that updates an initial model and enrollment model based on input embedding vectors, using confidence values and state information from sensors to enhance fake fingerprint detection, including adjustments to hit counts and data elimination to optimize model performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a fingerprint recognition system uses a pre-trained model with fixed training database, then the system structure is simple and easy to implement, but the system cannot adapt to changing environments and fails to distinguish fake fingerprints in new conditions
Solution Approach 1:
The patent implements dynamic model updating where the fingerprint recognition model is continuously adapted to new environments through incremental learning. The system updates the model parameters based on new fingerprint data while maintaining the core structure, allowing the system to adapt to changing conditions without complete retraining. This resolves the contradiction by making the system dynamically adaptable rather than statically fixed.
Solution Approach 2:
The patent performs preliminary actions by pre-training the model with initial fingerprint data before deployment, and then prepares the system for future adaptations by establishing the update mechanism in advance. The model is pre-configured with basic recognition capabilities and the framework for continuous learning is established beforehand, enabling smooth adaptation to new environments without ad-hoc complexity.
2Measurement precision
If the training database covers all possible use environments, then the measurement precision of fake fingerprint detection is high, but the data storage requirement and model complexity become excessive
Solution Approach 1:
The patent applies partial action by selectively updating the model with only the most relevant new fingerprint data rather than requiring complete coverage of all possible environments. The system processes and incorporates new data incrementally, focusing on the most impactful updates. This allows the system to improve detection accuracy progressively without needing to store and process excessively large volumes of training data for all conceivable scenarios.
Solution Approach 2:
The model dynamically adjusts its parameters based on incoming fingerprint data, allowing it to adapt to new environments on-demand rather than requiring pre-collection of all possible training data. This dynamic adaptation mechanism enables the system to achieve high detection precision in specific environments without maintaining a massive comprehensive training database.
3Adaptability or versatility
If the system continuously updates the model with new fingerprint data, then the adaptability to new environments improves, but the risk of overfitting and model instability increases
Solution Approach 1:
The patent implements feedback mechanisms where the system monitors the performance and stability of the updated model. The update process incorporates validation steps that check whether new data improves or degrades model performance. If overfitting is detected or stability deteriorates, the system adjusts the update process or rejects problematic data. This feedback loop maintains model reliability while enabling continuous adaptation to new environments.
Solution Approach 2:
The system prepares cushioning measures by implementing validation and verification steps before fully committing to model updates. New fingerprint data is evaluated and tested before being incorporated into the model, and rollback mechanisms are in place to restore previous stable model versions if updates cause instability. This beforehand cushioning protects against overfitting and model collapse while allowing adaptive updates.
Data Source
AI summary
A processor-implemented method includes: obtaining an input embedding vector corresponding to an input fingerprint image for authentication; determining a confidence value of the input embedding vector based on fingerprint data of an initial model including either one or both of a trained real fingerprint determination model and a trained fake fingerprint determination model that are provided in advance; and updating the initial model based on the input embedding vector, in response to the confidence value being greater than or equal to a first threshold.