Improving the Face Gender Classification by the Set of Features

Abstract

Gender recognition using face images is one of interesting for practical applications face analysis tasks.Most of the existing studies have focused on face images acquired under controlled conditions, such as famous FERET database. However, real applications require gender classification on real­ life faces, which is much more challenging due to significant appeamnce wriations in unconstrained scenarios.In this paper, we investigate gender recognition on real-life faces using the Labeled Faces in the Wid (LFW) dataset and our own RUS-FD dataset. We propose a gender classifier using three types of local features: Scale Inwriant Feature Transform (SIFT) which is one of the most commonly-used ones because it is invariant to image scaling, translation and rotation, Histogram of Oriented Gradient (HOG) features, which is able to capture local shape information from the gmdient structure with easiy controllable degree of invariance to translations and the Gabor wavelets which reflect the multi-scale directional information. We obtain the performance of 96.79% by applying boosting learning on LlW dataset 92.64% by applying Support Vector Machine (SVM) on RUS-FD dataset. The approach propolled in this paper h promising to be further studied on other face clusification tasks, such as age estimation and emotion recognition.

Keywords

• face analysis
• gender classification
• machine learning
• features