Parameter Optimization for Facial Recognition Feature Space
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing facial recognition techniques face challenges where class representative vectors of similar samples are mapped to close positions on the hypersphere, leading to incorrect classifications and degradation of feature space expression ability, resulting in reduced classification accuracy.
Innovation Solution
A parameter optimization method that extracts feature vectors, acquires classification results, and optimizes parameters based on classification errors and distance errors between class representative vectors to ensure non-overlapping feature areas in the feature space, using a gradient method to determine the position of class representative vectors and apply distance errors to the classification error for optimization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If class representative vectors are mapped close together on the hypersphere to improve feature space utilization, then the hypersphere is fully used, but similar samples are likely to be classified into wrong classes
Solution Approach 1:
The patent changes the parameter of class representative vector positions from close mapping to uniform distribution on the hypersphere surface. By modifying the positional parameters to achieve uniform distribution, the patent simultaneously improves classification accuracy (separating similar classes) while maintaining feature space utilization through comprehensive hypersphere coverage.
2Productivity
If class representative vectors of similar samples are mapped to close positions to improve learning efficiency, then learning becomes more efficient, but classification accuracy degrades due to overlapping feature areas
Solution Approach 1:
The patent applies equipotentiality by uniformly distributing class representative vectors on the hypersphere surface, creating equal spacing between different classes. This uniform distribution ensures that similar samples are adequately separated while maintaining balanced learning conditions across all classes, thereby improving both classification accuracy and learning efficiency.
Solution Approach 2:
The patent modifies the spatial parameters of class representative vectors from close positioning to uniform distribution. By changing the positional parameters to achieve maximum separation while maintaining hypersphere utilization, the patent resolves the contradiction between learning efficiency and classification accuracy.
3Reliability
If the hypersphere is not fully used to maintain clear class boundaries, then classification accuracy is maintained, but the expression ability of the feature space degrades
Solution Approach 1:
The patent changes the parameter of class representative vector distribution from sparse or non-uniform positioning to uniform distribution across the entire hypersphere surface. This parameter change enables full hypersphere utilization while maintaining clear class boundaries through equidistant positioning, thereby simultaneously improving feature space expression ability and classification accuracy.
Data Source
AI summary
A parameter optimization method includes extracting a feature vector using input data, acquiring a classification result of the feature vector and a class representative vector of every class serving as a classification target, and optimizing a parameter used in the extracting based on a classification error obtained using correct answer data and the classification result and a distance error between the class representative vectors such that areas of features of the classes in a feature space do not overlap each other.