Neural Network Authentication for ID Documents
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current identification systems using ID documents face challenges such as insecure data storage and transmission, large data volumes, and potential loss of personal data, especially when comparing encrypted data on smartphones for verification purposes.
Innovation Solution
Implementing a neural network-based system where a 3D data model is reduced to 128 digital reference features stored on an ID document, allowing secure comparison on a smartphone without exposing raw data, using a publicly known AI model to authenticate individuals by matching recorded 3D data with the document's data set.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If raw 3D data is stored on the smartphone for comparison, then identification accuracy is improved, but security is worsened due to plaintext storage and potential data loss
Solution Approach 1:
The patent extracts only the essential 128 feature values from the complete 3D head data and stores only these extracted features in the ID document. The smartphone processes the full 3D data through the neural network to generate comparison features, but never stores the raw data in plaintext. This extraction approach maintains identification accuracy while eliminating security risks associated with storing large volumes of sensitive personal data.
Solution Approach 2:
The neural network serves as an intermediary that transforms the raw 3D head data into a condensed 128-value feature representation. This intermediary processing layer allows the system to work with simplified data for storage and comparison purposes, reducing the attack surface for security vulnerabilities while preserving the essential identification information needed for accurate matching.
2Reliability
If complete 3D raw data is processed and stored, then identification reliability is improved, but data volume increases making storage difficult
Solution Approach 1:
The system extracts only the 128 most discriminative feature values from the complete 3D head data using a trained neural network. These 128 values capture the essential identification information while reducing the data volume from millions of 3D points to a manageable 128-byte representation that can be easily stored in the ID document's memory area.
Solution Approach 2:
The patent changes the parameter representation from raw 3D coordinate data (x, y, z positions for thousands of facial points) to a transformed feature space of 128 numerical values. This parameter transformation is achieved through neural network processing that identifies and encodes the most relevant facial characteristics, maintaining identification reliability while dramatically reducing storage requirements.
3Measurement precision
If AI model is updated to improve accuracy, then identification precision is improved, but device complexity increases
Solution Approach 1:
The system is designed to be dynamic in terms of model versioning. The ID document stores a version identifier for the neural network model used to generate the reference features. When the authentication device performs verification, it can download and load the corresponding model version to ensure consistency. This dynamic version management allows the system to improve identification precision through model updates without creating permanent complexity, as older versions can be replaced as needed.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The present disclosure relates to an authentication device (200) for authenticating a person (103) by means of an identification document (100) which has a storage area (110) in which a reference feature set (101) is stored, which has digital reference features (107) of a digital reference head model, wherein the reference feature set (101) comprises a response (212) using a trained neural network (205), with: an image acquisition device (201) which is configured to capture a head image (202) of the person;a processor (203) which is trained to generate a digital comparison head model (204) based on the recorded head image of the person, to provide a comparison feature set (206) with digital comparison features (207) as a response (210) of the trained neural network based on the digital comparison head model (204), and furthermore to determine a difference (208) between the digital reference features and the digital comparison features in order to authenticate the person.