Eye Detection via Geometric Segmentation and SVM Verification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional face recognition systems face challenges in accurately detecting eyes, especially under varying illumination conditions and when subjects wear eyeglasses, and often struggle to detect both eyes simultaneously.
Innovation Solution
The system divides an input face image into left and right images, detects eye candidates using geometric information, and verifies them using a support vector machine (SVM) algorithm, with a learning database to filter out non-eye candidates and determine successful eye detection based on a threshold value.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional eye detection methods (morphology operation, histogram analysis) are used on the entire face image, then the detection process is simple, but the method is sensitive to illumination variation and cannot accurately detect both eyes simultaneously
Solution Approach 1:
The face image is divided into multiple regions (left eye region, right eye region, and other regions) based on geometric constraints. The eye candidate detector searches only within the left and right eye regions, significantly reducing the search space and improving detection accuracy while being less sensitive to illumination variations in other face regions.
Solution Approach 2:
The system performs preliminary geometric constraint verification before final eye detection. The eye candidate evaluator uses geometric relationships (distance between eyes, position relative to face center, symmetry) to pre-filter candidate regions, eliminating obvious non-eye candidates before applying more complex detection algorithms.
2Measurement precision
If infrared strobe is used for eye position detection, then eye position can be detected, but afterimages are formed on eyeglass lenses and the infrared strobe is required
Solution Approach 1:
The system extracts and removes the harmful infrared strobe component from the detection process. Instead of using infrared illumination, the patent uses visible light imaging with geometric constraint-based detection, thereby eliminating the afterimage problem on eyeglass lenses while maintaining eye detection capability.
Solution Approach 2:
The patent introduces geometric constraints and region-based filtering as intermediary steps between image acquisition and eye detection. These intermediaries guide the detection process to focus on plausible eye regions, reducing false detections without requiring infrared illumination that causes afterimages.
3Measurement precision
If the search region for eye candidates is the entire face image, then no additional processing is needed, but the detection accuracy decreases and computation increases
Solution Approach 1:
The face image is segmented into specific eye candidate regions based on geometric constraints (left eye region, right eye region). The eye candidate detector searches only within these constrained regions rather than the entire face image, reducing computation time and improving detection accuracy by focusing on relevant areas.
Solution Approach 2:
Different regions of the face image are treated differently: the left and right eye regions are searched for eye candidates with specific geometric constraints, while other regions are excluded from eye candidate detection. This local differentiation improves efficiency by applying detection resources only where eyes are likely to be found.
Data Source
AI summary
An apparatus for and a method of detecting eyes from an input face image. The apparatus for detecting eyes includes: an eye candidate detector which divides an input face image into left and right images and detects at least one eye candidate from limited image regions of the left and right images; an eye candidate evaluator which evaluates the eye candidates by evaluating each combination of the eye candidates using geometric information as to the eye candidates to filter out eye candidates that cannot be eyes; a learning database which stores a plurality of face images in which positions of eyes are arranged and a plurality of face images which do not include eyes or in which positions of eyes are not arranged; and an eye candidate verifier which verifies the eye candidates with reference to the learning database and outputs an eye detection result signal.