Face Verification Using Light Masks and GMMs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional face verification technologies divide facial images into blocks, which increases learning and verification time and may not adequately account for the entire facial area, leading to inefficiencies and vulnerability to illumination variations.
Innovation Solution
A face verification system and method that extracts features from the entire facial area using light masks to create variously illuminated facial images, generating Gaussian Mixture Models (GMMs) for both user and non-user databases, and calculating log-likelihood values to verify user identity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If facial images are divided into blocks for feature extraction, then GMM learning can be performed excellently with a small number of images, but learning and verification time increase and the entire facial area may not be sufficiently taken into account
Solution Approach 1:
The patent applies segmentation by dividing the facial verification process into two distinct stages: (1) training phase using block-based GMM learning to capture local facial features efficiently, and (2) verification phase using holistic facial features to reduce verification time. This segmentation allows each stage to use the most appropriate feature representation for its specific purpose.
Solution Approach 2:
The patent dynamically switches between different feature extraction approaches depending on the operational phase. During training, it uses block-based features for comprehensive learning, while during verification, it transitions to holistic feature comparison for speed, creating a dynamic adaptation to operational requirements.
2Quantity of substance
If facial images are divided into blocks for feature extraction, then a plurality of data items can be obtained using a small number of images, but the entire facial image may not be sufficiently taken into account
Solution Approach 1:
The patent merges two complementary approaches: block-based feature extraction (which provides detailed local features and increases data quantity) and holistic facial feature extraction (which preserves overall facial structure information). By combining both methods in different phases, it achieves both data richness and structural integrity.
3Loss of information
If features are extracted from the entire facial area without block division, then the entire facial area is sufficiently taken into account, but the amount of data extracted through feature extraction decreases
Solution Approach 1:
The patent performs preliminary block-based GMM learning during the training phase to generate comprehensive feature data and establish baseline models. This preliminary action richly populates the feature space before the actual verification uses holistic features, ensuring both data quantity and structural information are captured at appropriate stages.
4Device complexity
If conventional face verification methods are used, then the process is simple, but the system is vulnerable to illumination variations
Solution Approach 1:
The patent changes the parameter representation by applying light masks that simulate various illumination conditions to the training images. This transforms the training data to include diverse lighting scenarios, enabling the GMM models to learn illumination-invariant features and improving robustness without significantly increasing verification complexity.
Data Source
Figure 1A~1B
Figure 2A~2C
Figure 3
AI summary
The present invention relates to a system and method for verifying the face of a user using a light mask. The system includes a facial feature extraction unit for extracting a facial feature vector from a facial image received from a camera. A non-user Gaussian Mixture Model (GMM) configuration unit generates a non-user GMM from a facial image stored in a non-user database (DB). A user GMM configuration unit generates a user GMM by applying light masks to a facial image stored in a user DB. A log-likelihood value calculation unit inputs the facial feature vector both to the non-user GMM and to the user GMM, thus calculating log-likelihood values. A user verification unit compares the calculated log-likelihood values with a predetermined threshold, thus verifying whether the received facial image is a facial image of the user.