Gender and Age Classification of Human Faces for Automatic Detection of Anomalous Human Behaviour

In this paper, we introduce an approach to classify gender and age from images of human faces which is an essential part of our method for autonomous detection of anomalous human behaviour. Human behaviour is often uncertain, and sometimes it is affected by emotion or environment. Automatic detection can help to recognise human behaviour which later can assist in investigating suspicious events. Central to our proposed approach is the recently introduced transfer learning. It was used on the basis of deep learning and successfully applied to image classification area. This paper is a continuous study from previous research on heterogeneous data in which we use images as supporting evidence. We present a method for image classification based on a pre-trained deep model for feature extraction and representation followed by a Support Vector Machine classifier. Because very few data sets with labels of gender and age exist of face images, we build one dataset named GAFace and applied our proposed method to this dataset achieving excellent results and robustness (gender classification: 90.33% and age classification: 80.17% accuracy) approaching human performance.