Gender classification in live videos

Human facial gender classification is an important task in live videos. However, it is still challenging in real applications due to motion blur, object occlusion and extreme illumination in real live videos. In this paper, we propose the Multi-Branch Voting CNN (MBV-CNN) framework which first detects and extracts the human face images in live videos, then apply adaptive brightness enhancement on each face image before feeding them into three CNN branches to settle the extreme illumination problem, and finally, we apply a majority voting scheme to reduce the influences from motion blur, object occlusion to further improve classification accuracy. Our method significantly outperforms the state-of-the-art solutions on the LFW dataset and our collected real-world live videos dataset called Gender Classification for Live Videos (GCLV), with respectively averaging 98.11% and 95.36% classification accuracy.

1. Introduction

With the rapid development of the Internet, live video applications have attracted massive users, which enables us to share daily life videos with others. These sharing actions continuously generate large amount of real-world videos and a great proportion of these videos mainly record human faces, so face analysis in videos is becoming more and more important in real application for video content inspection and recommendation. Gender classification is one of the most important video analysis tasks. As shown in Figure 1, classification tasks in live videos are more difficult than traditional videos. The difficulties mainly come from two aspects: 1) Most of live videos are captured by mobile devices rather than professional camera equipment, some are even captured with the devices in motion. These videos are affected by motion blur or object occlusion. 2) The locations of live video capturing are quite casual, such as bedroom, living room, market, outdoor and etc. This leads to great variations in lighting conditions with the problem of extreme illumination which is great challenge for gender classification. Fig. 1:

Face images in live videos dataset. These face images show the challenges of gender classification in real world. The face images in the first row show extreme illumination in live videos and the face images in second row suffer from motion blur or object occlusion.