Biometric Face Fraud Detection via Depth Map Variance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current facial recognition access control systems are vulnerable to 'plane attacks' where a two-dimensional representation of an authorized person's face is used to gain access, and existing fraud detection methods are not always effective, especially for semi-flat or curved frauds.
Innovation Solution
A method that captures at least two images of a face from different positions or angles, processes these images to determine a flatness score by calculating the variance of depth maps, and uses this score to detect fraud, thereby conditioning the identification process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If facial recognition systems use single image comparison for access control, then the system is simple and fast, but it becomes vulnerable to plane attacks using two-dimensional representations
Solution Approach 1:
The patent transitions from 2D image comparison to 3D depth map analysis. By capturing depth information and calculating variance of depth values, the system adds a dimensional aspect that inherently distinguishes flat photographs from three-dimensional faces, resolving the vulnerability to plane attacks while maintaining computational efficiency.
Solution Approach 2:
The patent changes the parameter being measured from 2D pixel intensity values to 3D depth values. By calculating the variance of depth values across facial regions, the system creates a new parameter (depth variance) that naturally differentiates between flat and three-dimensional surfaces, improving reliability without significantly increasing complexity.
2Reliability
If multiple images from different angles are captured to detect fraud, then fraud detection reliability improves, but the processing time and system complexity increase
Solution Approach 1:
The patent performs preliminary action by capturing multiple images and computing depth maps before the actual authentication decision. The depth variance calculation is prepared in advance, allowing the system to quickly compare pre-computed metrics rather than processing raw images in real-time, thus reducing processing time while maintaining high detection accuracy.
3Reliability
If eye blink detection is used to distinguish real faces from photographs, then some frauds are detected, but the method fails when photographs are taken for given periods or at specific positions
Solution Approach 1:
The patent replaces the mechanical/behavioral approach of eye blink detection with a geometric/structural approach using depth map analysis. Instead of relying on dynamic behavioral cues that can be controlled or absent, the system uses static geometric properties (depth variance) that fundamentally differ between flat and three-dimensional surfaces, making the detection method robust against various attack scenarios including posed photographs.
Data Source
Figure 1~2
Figure 3
Figure 4a~4b
AI summary
The invention relates to a method of detecting fraud for an access control system using biometric recognition, comprising the steps consisting in: taking at least two images (S2) of a face presented to the access control system by an individual (I) according to at least two positions of the individual in front of the access control system, or two angles of picture- taking of the individual, processing these two images to determine a score representative of the flatness of the face appearing on the images, and as a function of this flatness score, detecting a possible fraud on the part of the individual, said detection conditioning the implementation of an identification processing of the individual by the access control system. The invention also relates to an access control system implementing said method.