Person Recognition Threshold Setting via Similarity Distribution Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing person recognition systems face difficulties in setting an appropriate threshold value for determining whether a person is registered, as illumination and photography conditions vary, leading to challenges in maintaining a balance between false rejection and false acceptance rates.
Innovation Solution
A person recognition apparatus with a threshold value setting unit that calculates and adjusts the threshold based on similarity distributions between facial feature information from registered and non-registered individuals, using a mode selection unit to switch between operation and adjustment modes, allowing for easy determination of appropriate threshold values.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If a fixed threshold value is used for person recognition, then the system operation is simple, but the false rejection rate and false acceptance rate cannot be controlled within allowable ranges under varying illumination and photography conditions
Solution Approach 1:
The threshold value is made dynamic rather than fixed. The system automatically adjusts the threshold value based on the distribution of similarity measures calculated from facial feature information, allowing the threshold to adapt to varying illumination and photography conditions while maintaining reliable person recognition accuracy
Solution Approach 2:
The system performs self-adjustment of the threshold value without requiring manual intervention. By automatically calculating similarity distributions and determining appropriate threshold values, the system serves itself in optimizing its recognition performance under different environmental conditions
2Reliability
If the threshold value is manually adjusted for each installation location, then the false rejection and false acceptance rates can be controlled, but the setting process becomes complex and time-consuming
Solution Approach 1:
The system automatically determines appropriate threshold values without requiring manual adjustment. It calculates similarity distributions from facial feature information and self-adjusts the threshold to control false rejection and false acceptance rates, eliminating the need for complex manual setting processes at each installation location
Solution Approach 2:
The system changes the threshold parameter automatically based on calculated similarity distributions. By adjusting the threshold value dynamically according to the distribution characteristics of similarity measures, the system optimizes recognition performance without manual intervention
Data Source
AI summary
According to one embodiment, an apparatus includes input unit, detecting unit, extraction unit, storage, selection unit, determination unit, output unit, and setting unit. The selection unit selects operation or setting modes. In operation mode, it is determined whether captured person is preregistered person. In setting mode, threshold for the determination is set. The determination unit determines, as registered person and when operation mode is selected, person with degree of similarity between extracted facial feature information and stored facial feature information of greater than or equal to threshold. The setting unit sets, when setting mode is selected, threshold based on first and second degrees of similarity. First degree of similarity is degree of similarity between facial feature information of the registered person and the stored facial feature information. Second degree of similarity is degree of similarity between facial feature information of person other than registered person and stored facial feature information.