Hybrid classification for face recognition with virtual samples and ensemble neural networks

Abstract

This paper presents a novel hybrid classifier with artificially generated virtual training samples for face recognition. Our approach obtains a synergy effect by combining the results of two heterogeneous classifiers, and utilizes the nearest neighbor approach in feature angle space and a connectionist model. The first is a classifier called the nearest feature angle (NFA) based on angular information, which finds the most similar to the query from a given training set. The second is a classifier based on the recall of stored frontal projection of input features using a frontal recall network (FRN) that finds the most similar frontal one among the stored frontal feature set. For FRN, we used an ensemble neural network consisting of multiple MultiLayer Perceptrons (MLPs), each of which is trained independently to enhance generalization capability. Further, both classifiers used a virtual training set generated adaptively according to the spatial distribution of each person's training samples. Finally, the results of the two classifiers are combined to comprise the best matching class, and the corresponding similarity measure is used to make the final decision. The proposed classifier achieved an average classification rate of 96.33% against a large group of different sets of test images, and its average error rate is 61.2% of that of the nearest feature line (NFL) method, and achieves a more robust classification performance.