Simple hybrid classifier for face recognition with adaptively generated virtual data

Authors:
Yeon-Sik Ryu;Se-Young Oh
Affiliations:
Department of Electrical Engineering, Intelligent Systems Laboratory, Pohang University of Science and Technology (POSTECH), Pohang, Kyungbuk, South Korea;Department of Electrical Engineering, Intelligent Systems Laboratory, Pohang University of Science and Technology (POSTECH), Pohang, Kyungbuk, South Korea
Venue:
Pattern Recognition Letters
Year:
2002

Citing 4
Cited 3

Face recognition using a hybrid supervised/unsupervised neural network

Pattern Recognition Letters
Face Recognition: The Problem of Compensating for Changes in Illumination Direction

IEEE Transactions on Pattern Analysis and Machine Intelligence
Face Recognition: Features Versus Templates

IEEE Transactions on Pattern Analysis and Machine Intelligence
Face recognition using the nearest feature line method

IEEE Transactions on Neural Networks

Linear generalization probe samples for face recognition

Pattern Recognition Letters
Using the original and 'symmetrical face' training samples to perform representation based two-step face recognition

Pattern Recognition
Integrate the original face image and its mirror image for face recognition

Neurocomputing

Quantified Score

Hi-index	0.10

Visualization

Abstract

This paper presents a simple hybrid classifier for face recognition with artificially generated virtual training samples. Two sub-classifiers that work on eigenface space, use angular information obtained from training samples and the query feature point. First, training data set was expanded by adding virtual training samples generated adaptively according to the spatial distribution of each person's training samples. Second, a classifier, called the nearest feature angle (NFA) method, finds the most similar sample from an augmented training set to the query sample. Third, after finding the best matched feature line by applying the nearest feature line (NFL) method, the modified nearest feature line (MNFL) method finds the angular information between the query feature point and its projection onto best matched feature line. Finally, the hybrid classifier determines the class by comparing the angular information obtained by the two sub-classifiers. The proposed hybrid classifier exhibits an average error rate of 4.05%, which is 80.2% of that of the standard NFL method with improved robustness for different test sets of facial images.