Mobile Face Authentication Using TEE Vector Verification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Mobile devices with trusted execution environments (TEE) have limited computing power and memory capacity, making it challenging to perform secure facial authentication and recognition, which can be tampered with in the regular execution environment, exposing face recordings to security risks.
Innovation Solution
A method that utilizes a trained model in the regular execution environment to analyze facial images and generate a multi-dimensional vector, which is then processed in the TEE for verification by comparing it with a stored vector, ensuring secure authentication by leveraging the full computing power of the regular environment and the security of the TEE.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If facial authentication is performed in the trusted execution environment (TEE), then security is improved, but computing power and memory capacity are insufficient
Solution Approach 1:
The authentication system is divided into two segments: the TEE handles only security-critical operations (verification of multi-dimensional vectors and generation of authentication results), while the regular execution environment performs computationally intensive tasks (image analysis using trained models). This segmentation allows each environment to operate within its capability boundaries while maintaining overall security.
Solution Approach 2:
Multi-dimensional vectors serve as an intermediary data structure that bridges the regular execution environment and TEE. The vectors are generated in the regular environment from facial images, then transferred to the TEE for secure verification against stored template vectors. This intermediary approach enables secure authentication without requiring the TEE to perform full image analysis.
2Power
If facial authentication is performed in the regular execution environment, then computing power is sufficient, but security is compromised due to potential tampering
Solution Approach 1:
The security-critical verification function is extracted from the regular execution environment and placed in the TEE. Specifically, the verification of multi-dimensional vectors and the generation of authentication results are removed from the vulnerable regular environment and executed exclusively in the secure TEE, leaving only the less sensitive image analysis in the regular environment.
Solution Approach 2:
Instead of storing and comparing raw facial images in the TEE, the system creates a mathematical representation (multi-dimensional vector) of the facial features. This vector copy captures the essential authentication information while being more efficient for secure storage and comparison in the TEE environment.
3Measurement precision
If full facial image data is stored in the TEE, then authentication accuracy is improved, but memory capacity requirements exceed TEE limitations
Solution Approach 1:
The system transforms the facial image data from its original high-dimensional pixel format into a compressed multi-dimensional vector representation. This parameter transformation reduces the data size significantly while preserving the essential facial features needed for accurate authentication. The vector format enables efficient storage in the TEE's limited memory while maintaining authentication precision.
Data Source
AI summary
Disclosed herein is a method for authenticating a user of a mobile device. The method includes:a) imaging at least one first image of a face by using at least one camera of the mobile device;b) providing the at least one first image to a regular execution environment of a processor of the mobile device and providing the at least one first image to a trusted execution environment of the processor;c) analyzing the at least one first image using at least one trained model in the regular execution environment thereby determining a multi-dimensional vector comprising image information; andd) a face verification step.