Detection of sculpted faces on building facades

The work presents the first approach of face detection on building facades. Human sculptures are typical for a set of historic architectural styles. Thus the presence of faces can serve as an additional discriminative feature while performing image- based architectural style classification. The approach employs Viola-Jones face detector on a self-compiled image database of facades, featuring sculpted faces. At the first stage of the experimental setup OpenCV face detector, trained on real human faces, is used to detect the sculpted faces. Since the performance rate is unsatisfactory, the second stage of experiments is issued. Here new cascade classifiers are trained on the training image dataset of sculpted faces. The classifier with the best performance is chosen to carry out the third stage of the experimental setup on the testing dataset of images. The results indicate that the final detector achieves high detection rate on unfamiliar images. Furthermore, it considerably outperforms the OpenCV base face detector.