Adaptive Whitelist Determination Using Dual Transformation Formulas
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing face recognition systems face challenges in accurately determining whether a person is registered in a white list when photographing conditions change or the number of images for creating a subspace is small, leading to increased false alarms and instability in results.
Innovation Solution
A white list inside or outside determining apparatus that uses two transformation formulas, one created from preliminary learning images and another from application learning images, with adjustable weights based on the number of images, to perform matching and registration, ensuring stable and environment-adapted results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single transformation formula is used for face feature extraction, then the system is simple and fast, but recognition accuracy deteriorates when photographing conditions change
Solution Approach 1:
The system dynamically switches between multiple transformation formulas based on the number of available learning images. When fewer than a threshold number of learning images are available, it uses the first transformation formula (simple, fast). When more images are available, it uses the second transformation formula (more accurate but complex). This dynamic adaptation resolves the contradiction between simplicity and accuracy.
Solution Approach 2:
The system changes the parameter of transformation formula selection based on the number of learning images. By using the number of learning images as a threshold parameter, the system switches between different transformation formulas (first and second formulas) to adapt to varying data availability conditions, thereby maintaining both simplicity and accuracy as needed.
2Measurement precision
If more learning images are used to create subspace, then recognition accuracy improves, but the time required for learning and processing increases
Solution Approach 1:
The system dynamically adjusts the amount of learning data used based on availability. When fewer images are available, it processes only those (faster). When more images are available, it uses them to improve accuracy. This dynamic data selection resolves the time-accuracy tradeoff.
Solution Approach 2:
The system applies partial action by using only the necessary number of learning images available, rather than requiring a fixed large number. It can achieve acceptable accuracy with partial data when time is constrained, and uses excessive data (more images) only when time permits for improved accuracy.
3Adaptability or versatility
If the same transformation formula is used for all environments, then the system is consistent, but it cannot adapt to different photographing conditions
Solution Approach 1:
The system dynamically selects transformation formulas based on environmental conditions (number of learning images available). This dynamic selection enables adaptation to different environments without requiring a completely complex system, as it simply switches between pre-defined formulas based on data availability.
Solution Approach 2:
The system prepares multiple transformation formulas in advance (first and second formulas) and selects the appropriate one based on the number of learning images. This preliminary preparation allows the system to adapt to different environments without complex real-time computation, resolving the adaptability-complexity contradiction.
Data Source
AI summary
A white list inside or outside determining apparatus includes: a first feature data extracting unit which extracts first feature data from an image by using a first transformation formula created based on preliminary learning images; a second feature data extracting unit which extracts second feature data from an image by using a second transformation formula created from the preliminary learning images and application learning images; a first matching unit which performs matching between a registration image and a collation image by using the first transformation formula; and a second matching unit which performs matching between a registration image and a collation image by using the second transformation formula. Weights of a matching result of the first matching unit and a matching result of the second matching unit are changed according to the number of preliminary learning images and the number of application learning images.