Face Recognition Using Boundary Samples and SVDD
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing face recognition methods face challenges in achieving high accuracy during image searches due to variations in face orientation, expression, and illumination, especially when the number of registered face images is small, leading to erroneous recognition and increased processing complexity.
Innovation Solution
An image recognition apparatus and method that acquires and extracts feature amounts from object images, registers training samples near the class boundary, calculates reliability for each class using Histogram Intersection Kernel (HIK), and performs identification based on these reliabilities to enhance recognition accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If multiple face images are registered to improve recognition accuracy under varied conditions, then recognition reliability improves, but processing complexity and amount of data increase
Solution Approach 1:
The patent extracts only the essential boundary information from multiple face images by identifying support vectors that define the class boundaries. Instead of processing all registered face images, the system extracts and stores only the critical boundary-defining samples, significantly reducing processing complexity while maintaining recognition reliability.
Solution Approach 2:
The patent transforms the recognition approach by changing from direct image comparison to using derived parameters (support vectors and boundary information). This parameter transformation allows the system to maintain high recognition accuracy while working with a compressed representation of the training data, reducing processing requirements.
2Loss of time
If the number of registered face images is small, then processing time is reduced, but recognition accuracy deteriorates due to insufficient training samples
Solution Approach 1:
The patent performs preliminary action by pre-identifying and storing the boundary-defining support vectors during the registration phase. When a recognition query is received, the system only needs to compare against these pre-identified boundary samples rather than processing all training images, enabling fast processing even with limited training data while maintaining accuracy through the use of critical boundary information.
3Adaptability or versatility
If partial spaces with large expanse are used to accommodate varied capture conditions, then adaptability improves, but similarity between partial spaces decreases leading to erroneous recognition
Solution Approach 1:
The patent applies local quality by focusing on the specific boundary regions that define class separation rather than using large, uniform partial spaces. The support vectors identify local boundary characteristics that are critical for differentiation, allowing the system to maintain high measurement precision for similarity comparison while still accommodating variations in capture conditions through the boundary-focused approach.
4Measurement precision
If SVDD is applied to extract support vectors near identification boundary, then recognition accuracy improves, but the ability to determine reliability belonging to a category is lost
Solution Approach 1:
The patent introduces an intermediary mechanism that uses the support vectors and boundary information not only for accurate classification but also for computing reliability scores. The system calculates the distance from the input image to the class boundaries defined by support vectors, using this distance as a measure of confidence or reliability in the classification, thus recovering the reliability information that would otherwise be lost.
Data Source
AI summary
There are provided an image recognition apparatus and an image recognition method according to which, when registering a face model for a face image, face images are sequentially acquired from frames of a moving image, feature amounts are extracted from the sequentially acquired face images, and a hypersphere defined by Support Vector Data Description (SVDD) is obtained from the group of training samples. Then, training samples in a boundary region of the hypersphere are registered as boundary samples. When performing matching of a face image of a recognition target, the reliability is calculated using the boundary samples, and a person with the highest reliability is output as a recognition result.