Facial Recognition Morphing Attack Detection via ID Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing facial recognition systems are inadequate in detecting morphing attacks, where a facial image is manipulated to combine characteristics of two individuals, leading to fraudulent identification, and struggle to differentiate between authentic and manipulated images, especially when angle differences in characteristic points are minimized.
Innovation Solution
Incorporating additional ID card data, such as validity date, age, and gender, into the comparison algorithm to enhance the similarity measure, and using machine learning algorithms like convolutional neural networks to improve the accuracy of facial recognition, while also considering age-dependent and gender-related changes in facial features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional facial recognition algorithms are used to compare facial images, then basic identification can be performed, but the system cannot detect morphing attacks that blend characteristics of two individuals
Solution Approach 1:
The patent segments the facial image analysis into multiple independent components: detecting characteristic points (eyes, nose, mouth), calculating angles between these points, and comparing angular relationships separately from facial feature matching. This segmentation allows the system to detect morphing attacks by analyzing geometric relationships independently, improving detection accuracy without requiring complete algorithm redesign
Solution Approach 2:
The patent introduces a new dimension of analysis by calculating angles between characteristic points on the facial image, adding angular relationship comparison as an additional layer beyond traditional facial feature matching. This dimensional expansion enables the system to detect morphing attacks that preserve facial features but alter geometric relationships, thereby improving reliability
2Reliability
If the angle method is used to detect morphing attacks by comparing characteristic point angles, then detection capability improves, but sophisticated morphed images with minimized angle differences can evade detection
Solution Approach 1:
The patent merges multiple detection methods into a unified system: traditional facial feature matching is combined with angular relationship analysis, and both are integrated into a single assessment that produces a similarity measure. This combination allows the system to detect morphing attacks through angular analysis while maintaining robust facial recognition, overcoming the limitation of angle-based methods alone
Solution Approach 2:
The patent changes the assessment parameters by introducing angular relationships as new comparison metrics alongside traditional facial feature distances. By calculating angles between characteristic points and comparing these angular relationships, the system transforms the detection approach from purely metric-based to include geometric relationships, thereby improving detection precision against sophisticated morphing attacks
3Ease of operation
If only facial images are used for identification, then the process is simple, but the system cannot verify the validity period or detect fraudulent use of expired documents
Solution Approach 1:
The patent makes the identification system multi-functional by integrating not only facial recognition but also document validation, expiration date checking, and fraud detection capabilities. The computer system performs multiple functions: comparing facial images, verifying ID card validity periods, assessing morphing attacks, and producing comprehensive authentication results, thereby enhancing reliability while maintaining operational simplicity through automation
Data Source
Figure 1
Figure 2
Figure 3
AI summary
In a procedure for authenticating a person using facial images and/or for detecting manipulation of facial images, an identity document facial image (2) from an identity document (3) of a person to be identified (4) is used. A computer system (7) assesses whether and to what extent the identity document facial image (2) and a comparison facial image (5) correspond. The computer system (7) includes a comparison algorithm (8), which compares the identity document facial image (2) and the comparison facial image (5). This comparison forms the basis of the assessment, which is represented by a similarity measure (13). In addition to the identity document facial image (2), further identity document data (9) from the identity document (3) are used, whereby the further identity document data (9) are used by the computer system (7) and can influence a value of the similarity measure (13).The other identification data (9) include a validity date and are read electronically from an integrated circuit of the identification document (3), in particular by means of radio.