Facial image retrieval using hybrid classifiers

This paper considers hybrid classification architectures for contents based facial image retrieval using gender, ethnic origin and identity as retrieval cues. The hybrid approach consists of an ensemble of RBF networks and inductive decision trees (DT). Experimental cross validation (CV) results on a collection of 3006 frontal face images from the FERET database (corresponding to 1009 subjects) yield-(a) an average accuracy rate of 96% 94% and 92% on the gender, ethnic and identity tasks respectively when their corresponding classifiers were used individually and (b) an average cumulative accuracy rate of 86% for the case, when the ethnic and identity classifiers were used jointly and 83% when all the three classifiers were used together. The benefits of our hybrid architecture, beyond the high accuracy achieved, include (i) robustness via query by consensus provided by the ensembles of RBF networks, and (ii) flexible and adaptive thresholds as opposed to ad hoc and hard thresholds provided by DT