SVM Feature Vector Removal for Face Recognition Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional face detection and recognition methods, such as SVM, face challenges with low recognition accuracy when the sample set is small due to data overfitting, resulting in high false positives and false negatives.
Innovation Solution
The method involves optimizing the SVM model by selecting and removing feature vectors corresponding to false positives, retraining the model, and adjusting kernel parameters to achieve a higher ratio of sample number to support vector number, thereby increasing recognition accuracy and reducing false negatives.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional SVM methods are used with small sample sets, then the model can be trained quickly, but recognition accuracy is low due to data overfitting and high false positives
Solution Approach 1:
The patent extracts and removes false positive feature vectors from the training set through iterative testing and identification. By systematically identifying and removing FP feature vectors that cause overfitting, the method improves recognition accuracy while maintaining model reliability with small sample sets
Solution Approach 2:
The patent changes the composition of the training set by dynamically adjusting the feature vector set through iterative removal of false positives. This parameter change in the training data composition allows the SVM model to achieve better generalization with small sample sets, reducing both false positives and improving accuracy
2Loss of information
If more feature vectors are retained in the training set, then more information is preserved, but the ratio of sample number to support vector number decreases leading to overfitting
Solution Approach 1:
The patent implements a feedback mechanism through iterative testing where the model performance is continuously evaluated and used to guide the removal of false positive feature vectors. This feedback loop ensures that information is retained only from high-quality feature vectors that improve rather than harm recognition accuracy
Solution Approach 2:
The patent applies partial action by selectively removing only the false positive feature vectors while retaining the majority of useful feature vectors. This selective approach maintains sufficient information for accurate recognition while preventing overfitting by removing only the harmful excess
3Measurement precision
If the sample set size is increased to improve recognition accuracy, then more data is available for training, but the processing time and computational complexity increase significantly
Solution Approach 1:
The patent extracts only the essential false positive feature vectors that are causing overfitting, rather than processing or analyzing all feature vectors exhaustively. This selective extraction approach improves accuracy by removing harmful elements while minimizing the computational time required for the optimization process
Data Source
AI summary
A computer-implemented method includes selecting a kernel and kernel parameters for a first Support Vector Machine (SVM) model, testing the first SVM model on a feature matrix T of n feature vectors of length m to produce false positive (FP) data set and false negative (FN) data set by a computer processor, wherein n and m are integer numbers, automatically removing feature vectors corresponding to the FP data set from the feature matrix T by the computer processor to produce a feature matrix T_best, retraining the first SVM model on the feature matrix T_best to produce a second SVM model, and checking if a ratio (T_best sample number)/(SVM support vector number) is above a threshold for the second SVM model on T_best. If the ratio is above the threshold, SVM predictions are performed using the second SVM model on the feature matrix T_best.