Face Recognition Negative Sample Generation for Fraud Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The imbalance in the quantity of positive and negative samples in face recognition systems leads to inaccurate face recognition models, as there are many positive samples but few negative samples, making it difficult to distinguish between genuine and fraudulent authentication attempts.
Innovation Solution
A method to generate negative samples by embedding positive samples in templates simulating real-life attack scenarios, including geometric distortions and reflective effects, to create a more authentic negative sample set for training face recognition models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If a large quantity of positive samples is used for training, then the face recognition model has sufficient genuine authentication data, but the negative samples remain insufficient leading to inaccurate fraud detection
Solution Approach 1:
The patent applies copying by generating synthetic negative samples through geometric transformations (rotation, scaling, flipping) and adding reflective effects to existing positive samples. This creates artificial attack samples that replicate real-world fraudulent authentication scenarios, thereby increasing the quantity of negative samples without requiring additional real attack data.
Solution Approach 2:
The patent segments the sample generation process into distinct transformation operations (geometric transformations, reflective effect additions, template embeddings). Each transformation is applied independently to generate diverse negative samples, allowing systematic expansion of the negative sample set while maintaining control over the characteristics of generated samples.
2Reliability
If more negative samples are obtained from real attack attempts, then the model accuracy improves, but the complexity of data collection and processing increases significantly
Solution Approach 1:
The patent applies self-service by enabling the system to generate its own negative samples from existing positive samples through automated geometric transformations and reflective effect additions. This eliminates the need for manual collection and processing of real attack data, as the system autonomously creates synthetic negative samples that capture fraud characteristics.
Solution Approach 2:
The patent applies parameter changes by systematically modifying sample parameters through geometric transformations (rotation angles, scaling factors, flip directions) and reflective effect parameters (reflection angles, transparency levels). These parameter variations generate diverse negative samples from a single positive sample, reducing data collection complexity while maintaining sample diversity.
3Ease of manufacture
If simple negative samples are used, then the training process is faster and easier, but the model fails to distinguish complex fraudulent authentication attempts
Solution Approach 1:
The patent applies composite materials by combining multiple transformation operations (geometric transformations + reflective effects + template embeddings) to create composite negative samples. These samples integrate multiple fraud characteristics simultaneously, enabling the model to learn complex fraudulent patterns while maintaining relatively simple generation processes through automated composition of transformation steps.
Data Source
Figure 1A~1B
Figure 2~3
Figure 4~5
AI summary
This application discloses a method and apparatus of generating a negative sample for face recognition, and a computer device. The method includes: obtaining a positive sample from a training sample library required for machine learning of face recognition; embedding the obtained positive sample in a selected negative sample template, to obtain an intermediate sample that simulates displaying of the positive sample in a display region of the negative sample template; and fusing the intermediate sample in a selected scenario sample, to obtain a negative sample required for machine learning of face recognition.