Self-Supervised Face Quality Recognition via Live-Image Similarity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for face image quality recognition in eKYC processes are inefficient due to extensive manual labeling required for binary classification, which is time-consuming and not effective in assessing the quality of ID images accurately.
Innovation Solution
A self-supervised face quality recognition method that pairs ID face images with live face images to generate similarity scores, allowing for the training of a multiclass classification or regression model without manual labeling, enabling accurate quality ranking and classification of ID face images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If binary classification model is used for face image quality recognition, then the model can output two classes (acceptable/unacceptable quality), but extensive manual labeling is required which is time-consuming
Solution Approach 1:
The system uses live face images as self-generated training labels by comparing them with ID face images through similarity scoring. This self-service mechanism eliminates the need for external manual labeling, allowing the model to train automatically using the eKYC system's own operational data.
Solution Approach 2:
The system performs preliminary similarity scoring between live and ID face images before model training. By pre-processing and ranking images based on similarity scores, the system prepares labeled training data in advance, reducing the time required during actual model training and deployment.
2Measurement precision
If binary classification is used, then the process is simpler, but the quality assessment accuracy is reduced compared to multiclass classification
Solution Approach 1:
The system segments the quality assessment into multiple discrete quality levels (e.g., top 10%, middle 10%, bottom 10% percentiles) rather than a single binary classification. This segmentation allows for more precise quality measurement while maintaining manageable model complexity through hierarchical or staged processing.
Solution Approach 2:
The system transitions from a one-dimensional binary classification to a multi-dimensional quality spectrum by introducing similarity scores as an additional dimension. This allows the model to assess quality along multiple axes (similarity score, ranking percentile, quality level) simultaneously, improving measurement precision without proportionally increasing complexity.
3Reliability
If manual labeling is performed extensively to achieve high accuracy, then model accuracy improves, but productivity decreases
Solution Approach 1:
The system generates its own training labels automatically by computing similarity scores between live and ID face images. This self-service approach eliminates the need for human annotators, maintaining high model accuracy through consistent automated labeling while dramatically improving training productivity and throughput.
Solution Approach 2:
The system implements feedback loops where similarity scoring results from actual eKYC operations are fed back into the training process. This continuous feedback mechanism allows the model to learn from real-world data patterns, improving accuracy over time while maintaining high productivity through automated iterative training.
Data Source
AI summary
Disclosed are computer-implemented methods, non-transitory computer-readable media, and systems for identity document face image quality recognition. One computer-implemented method includes pairing, for each user of a plurality of users and to form a pair of face images, an identity document (ID) face image and a live face image. For each pair of face images and based on a face similarity between the ID face image and the live face image, a similarity score for the ID face image is generated. Based on ID face images and similarity scores corresponding to the ID face images, a model for ID face image quality recognition is trained.