A Study of Multi-view Gender Recognition on a Large Database

In this paper, we conduct a systematic study of multi-view gender recognition on a large database. We focus on the basic methodologies of dealing with view point variations, and compare their performance on gender recognition problems. Experimental evaluation is conducted on a large multi-view database consisting of 23136 facial images of 241 subjects under 32 different views and 3 illuminations. We also introduce a Pyramid-SVM approach for facial recognition tasks. As an extension to the conventional “holistic features + SVM” framework, our proposed approach manages to preserve local information of face images that would otherwise be discarded in the holistic features. Experimental results demonstrate that the Pyramid-SVM approach significantly outperforms the traditional SVM classifier.