Facial Identification System Using Difference Feature Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional facial image identification systems face performance deterioration due to environmental changes, such as differences in gender, region, and lighting conditions, leading to increased costs when separate identifiers are generated for each environment.
Innovation Solution
A facial image identification system that compares feature amounts from training data and operational data to extract difference features, using machine learning to construct a secondary identifier for improved performance in varying environments, while limiting the cost by focusing on identifying only the difference features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If separate identifiers are prepared for each environment, then identification precision is improved, but cost increases
Solution Approach 1:
The patent segments the identification task into two parts: a general identifier for common features and a specialized identifier for environment-specific difference features. This segmentation allows the system to achieve high identification precision in different environments without preparing completely separate identifiers for each environment, thereby controlling costs while maintaining accuracy.
Solution Approach 2:
The patent applies local quality by creating specialized identifiers only for specific difference features that vary by environment (such as eye shape, nose size, lip shape) rather than creating entirely separate identifiers for all features. This localized approach improves identification precision for environment-specific variations while avoiding the high cost of comprehensive separate identifier systems.
2Quantity of substance
If a single identifier is used across all environments, then cost is reduced, but identification precision deteriorates
Solution Approach 1:
The general identifier serves multiple environments and feature types, providing universal identification capability for common facial features across different ethnic groups and environments. This multi-functionality reduces costs by avoiding duplicate identifier systems while maintaining adequate precision for non-difference features.
Solution Approach 2:
The patent introduces difference feature extraction as an intermediary process that identifies specific features requiring environment-specific handling. This intermediary mechanism allows the system to use the economical general identifier for most features while selectively applying specialized identifiers only where needed, balancing cost and precision.
3Adaptability or versatility
If identifiers are constructed from diverse environmental data, then adaptability is improved, but manufacturing complexity increases
Solution Approach 1:
The patent extracts only the necessary difference features that vary by environment (such as eye shape, nose size, lip shape) rather than processing all facial features. This extraction approach improves environmental adaptability by focusing on relevant variations while reducing the complexity of identifier construction by eliminating unnecessary processing of invariant features.
Solution Approach 2:
The patent performs preliminary analysis to identify difference features before constructing specialized identifiers. This preliminary action simplifies the overall construction process by pre-determining which features require environment-specific handling, thereby improving adaptability without proportionally increasing construction complexity.
Data Source
AI summary
A facial image identification system is provided to calculate a first feature amount related to each of features from first facial image data used in machine learning for constructing a first identifier that identifies the features of a face, calculate a second feature amount related to each feature from second facial image data obtained in the environment using the first identifier, and compare the first and the second feature amounts for each feature. The system extracts a feature having a large difference between the first and the second feature amounts as a difference feature, and constructs, by implementing machine learning that third facial image data is used, a second identifier for identifying the difference feature and/or a difference-related feature. The third facial image data is acquired in the environment that uses the first identifier and reflects the difference feature and/or the difference-related feature in the third facial image data.