Liveness Detection Using Eye Openness Confidence Scores
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current liveness detection methods fail to accurately differentiate between live users and spoofing attempts, particularly with high-definition video playback, leading to unreliable authentication transactions.
Innovation Solution
A method and system that calculate eye openness measures and confidence scores from captured authentication data, using detection windows and feature values to detect eye blinks, thereby determining user liveness by comparing maximum confidence scores against a threshold.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If known liveness detection methods are used, then authentication transactions can be conducted, but spoofing attempts using high definition video playback cannot be detected
Solution Approach 1:
The patent segments the eye region into multiple detection windows (first detection window and second detection window) positioned at different locations within the eye region. Each detection window independently calculates feature values and confidence scores, allowing the system to analyze different portions of the eye separately. This segmentation enables more precise detection of eye blink patterns while maintaining manageable computational complexity through localized analysis.
Solution Approach 2:
The patent introduces a temporal dimension by analyzing eye openness measures across multiple frames and calculating confidence scores based on temporal patterns of eye blinking. The system detects eye blinks by monitoring changes in eye openness over time rather than analyzing a single static image, adding the time dimension to the detection process. This temporal analysis significantly improves reliability in distinguishing live users from spoofing attempts.
2Reliability
If eye blink detection is implemented to verify user presence, then spoofing detection improves, but the system must process multiple detection windows and calculate confidence scores increasing computational complexity
Solution Approach 1:
The patent divides the eye region into multiple discrete detection windows, each independently processing feature extraction and confidence score calculation. This segmentation allows parallel processing of different eye regions, improving verification accuracy through comprehensive coverage while managing computational load by localizing processing to specific regions rather than analyzing the entire image globally.
Solution Approach 2:
The patent focuses computational resources on specific critical regions (the eye region and its sub-regions) rather than processing the entire image. By concentrating analysis on the detection windows within the eye region where blink patterns are most informative, the system achieves high verification accuracy with reduced overall computational complexity compared to full-image analysis.
3Measurement precision
If multiple detection window types and positions are used to calculate feature values, then eye blink detection accuracy improves, but the time required for authentication increases
Solution Approach 1:
The patent pre-defines specific detection window types and positions within the eye region based on anatomical landmarks and expected blink pattern locations. By segmenting the analysis into predetermined windows rather than searching all possible positions, the system achieves high measurement precision through targeted analysis while reducing authentication time by eliminating exhaustive search procedures.
Solution Approach 2:
The patent performs preliminary identification and positioning of detection windows within the eye region before the actual authentication transaction. Detection windows are pre-configured based on eye region geometry and expected blink characteristics, so that during authentication, the system can immediately begin feature extraction and confidence score calculation without time-consuming setup, thus achieving high precision efficiently.
Data Source
Figure 1~3
Figure 4~8
Figure 9~11
AI summary
A method determining user liveness is provided that includes calculating, by a device, eye openness measures for a frame included in captured authentication data, and storing the eye openness measures in a buffer of the device. Moreover the method includes calculating confidence scores from the eye openness measures stored in the buffer, and detecting an eye blink when a maximum confidence score is greater than a threshold score.