Biometric Detection via ROI Segmentation and Median Displacement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for detecting biometric characteristics in images, particularly facial features, are inefficient and costly due to difficulties in accurately determining the location of eyes and mouth, which hinders effective liveness detection and user authentication.
Innovation Solution
A system and method utilizing a smartphone or similar device with face tracker applications to generate a detection window around captured facial images, translating regions of interest to improve accuracy in calculating feature values, and employing tree-based models to determine precise eye locations through incremental positioning of detection windows.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional liveness detection methods are used with 2D pictures, then impostor detection is difficult, but accurate biometric characteristic detection is also difficult
Solution Approach 1:
The patent segments the facial image into multiple regions of interest (ROIs) including eyes, nose, and mouth. By dividing the face into distinct segments and analyzing each separately, the system can more accurately detect biometric characteristics and distinguish between real faces and photos. Each ROI is processed independently to extract specific features that contribute to liveness detection.
Solution Approach 2:
The patent replaces traditional mechanical/image-based liveness detection with depth sensing technology. By using a depth camera or structured light system to capture three-dimensional facial data, the system can detect subtle depth variations that distinguish real faces from flat photos, eliminating the need for complex 2D image analysis and manual feature location.
2Measurement precision
If accurate facial feature detection is performed manually, then liveness detection accuracy improves, but processing time and cost increase
Solution Approach 1:
The patent performs preliminary detection by first identifying the overall facial region and then systematically searching for specific features like eyes within predefined search areas. This hierarchical approach with predetermined search regions and stepwise refinement allows the system to quickly narrow down feature locations without exhaustive analysis, reducing processing time while maintaining accuracy.
Solution Approach 2:
The patent implements self-service through automated feature detection algorithms that independently locate and identify facial features without manual intervention. The system uses machine learning models and pattern recognition to automatically detect eyes, nose, and mouth positions, eliminating the need for time-consuming manual annotation while achieving high precision in feature localization.
3Quantity of substance
If detection window is arranged about captured image, then all facial characteristics appear in window, but accurate location of features cannot be determined
Solution Approach 1:
The patent divides the detection window into multiple smaller regions of interest, each focused on a specific facial feature such as eyes, nose, or mouth. By segmenting the window and analyzing each region separately with specialized detection algorithms, the system can capture all facial characteristics while simultaneously achieving high precision in locating each specific feature within its designated region.
Solution Approach 2:
The patent applies different detection strategies and parameter settings to different regions of the detection window based on local characteristics. Each region is optimized for detecting specific features (e.g., eye detection parameters for eye regions, nose detection parameters for nose regions), allowing accurate location determination while maintaining comprehensive capture of all facial characteristics across the entire window.
Data Source
Figure 1~3
Figure 4~5
Figure 6~9
AI summary
A method for detecting biometric characteristics in a captured biometric data image is provided that includes determining, by a processor, an approximate location for a biometric characteristic in a frame included in captured biometric data, and determining region of interest positions over the frame. Moreover, the method includes calculating a set of feature values for each position, generating a displacement for each set of feature values and generating a median displacement, and adjusting the biometric characteristic location by the median displacement.