Adaptive Facial Imagery Clustering via Spectral Clustering and Reinforcement Learning

In an era of big data, face images captured in social media and forensic investigations, etc., generally lack labels, while the number identities (clusters) may range from a few dozen to thousands. Therefore, it is practical importance cluster large unlabeled into efficient or even exact identities, which can avoid image labeling by hand. Here, we propose adaptive facial imagery clustering that involves representations, spectral clustering, reinforcement learning (Q-learning). First, use deep convolutional neural network (DCNN) generate adopt model construct similarity matrix achieve partition. Then, internal evaluation measure (the Davies–Bouldin index) evaluate quality. Finally, Q-learning as feedback module build dynamic multiparameter debugging process. The experimental results on ORL Face Database show effectiveness our method terms optimal clusters 39, almost actual 40 clusters; 99.2% accuracy. Subsequent studies should focus reducing computational complexity dealing with more images.