Dealing with inaccurate face detection for automatic gender recognition with partially occluded faces

Abstract

Gender recognition problem has not been extensively studied in situations where the face cannot be accurately detected and it also can be partially occluded. In this contribution, a comparison of several characterisation methods of the face is presented and they are evaluated in four different experiments that simulate the previous scenario. Two of the characterization techniques are based on histograms, LBP and local contrast values, and the other one is a new kind of features, called Ranking Labels, that provide spatial information. Experiments have proved Ranking Labels description is the most reliable in inaccurate situations