Passive Liveness Detection via Multi-Region Neural Embeddings
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing biometric systems face challenges in distinguishing between live and presentation attacks, particularly with passive liveness detection, where users do not need to perform specific actions, making it difficult to prevent unauthorized access.
Innovation Solution
The method involves using neural networks to generate embeddings from input images and cropped images focusing on facial regions, combining these embeddings to predict liveness, and denying access based on the prediction, with the neural networks trained on a dataset indicating live or presentation attack scenarios.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If passive liveness detection is implemented without user interaction, then user experience is improved, but the ability to detect presentation attacks deteriorates
Solution Approach 1:
The system segments the facial image into multiple regions of interest (eyes, nose, mouth, contours) and processes each region separately through dedicated neural network branches. This segmentation allows the system to capture subtle regional characteristics that indicate liveness without requiring user interaction, resolving the contradiction between passive operation and detection reliability.
Solution Approach 2:
The system transitions from analyzing single 2D images to processing multiple cropped regions and generating embeddings in a higher-dimensional feature space. By combining embeddings from different facial regions and using 3D facial landmark data, the system creates multi-dimensional representations that enhance presentation attack detection while maintaining passive operation.
2Measurement precision
If multiple cropped images and neural networks are used, then liveness detection accuracy is improved, but computational complexity increases
Solution Approach 1:
The complex detection task is segmented into multiple specialized neural networks, each processing specific facial regions (eyes, nose, mouth, contours). This modular segmentation improves detection accuracy for each region while allowing independent optimization and efficient computation, managing the overall system complexity.
Solution Approach 2:
The system extracts only the essential features from each facial region by generating compressed embeddings that capture the most discriminative characteristics. This extraction process reduces the dimensionality of processed data while preserving critical liveness indicators, improving accuracy without proportionally increasing computational complexity.
3Measurement precision
If facial region cropping is performed, then focus on relevant features is improved, but information from full context is lost
Solution Approach 1:
The system segments the face into multiple cropped regions (eyes, nose, mouth, contours) and processes each region separately. This segmentation focuses computational attention on relevant facial features while the concatenation of multiple regional embeddings preserves comprehensive facial information, balancing feature detection precision with information retention.
Solution Approach 2:
The system merges the embeddings from multiple cropped facial regions by concatenation, combining the focused regional features back into a comprehensive representation. This merging process recovers the holistic facial context that would be lost in individual crops while maintaining the precision benefits of region-specific analysis.
Data Source
AI summary
Methods, systems, and storage media are disclosed for passive liveness detection using artificial intelligence. Example implementations may receive an image of a person's face; generate a cropped version of that image; generate two different embeddings using two convolutional neural networks that are fed the image and the cropped image, respectively; generate a combined embedding that is a concatenation of the two embeddings; and generate, based on the combined embedding, an output indicating whether the facial portion corresponds to a live person. In addition, systems, devices, and methods for multi-factor authentication for transaction processing are provided. A point-of-transaction device captures customer information, biometric data, and images of identification documents and transmits the information to a transaction information server which receives the transaction request, queries one or more storage records to confirm the identity of the customer to the transaction and to determine whether the customer is authorized to engage in the transaction. The point-of-transaction device communicates a transaction identifier code and at least a portion of the transaction request to a transaction authority. The transaction authority transmits a confirmation signal to the point-of-transaction device based on the transaction identifier code and the transaction request.