Low-cost system for gender recognition using convolutional neural network

Abstract

Gender recognition of human face images is an important task in computer vision. The characters with the greatest gender diversity are the face and the pelvis, so the article uses face images to determine the gender. There are many reasons to automatically determine gender. One of them is visual surveillance. Other applications includes marketing, intelligent user interfaces, demographic studies. This paper presents a modern approach in identifying gender by using drones and specialized neural networks. This paper uses UAV data, it is a low cost data acquisition solution. The data has a very high resolution, so it is possible to obtain face cut-outs. Face cut-outs are then used to determine gender. The convolutional neural network AlexNet is used for classification. The system does not require any pre-processing and features extraction before classification. The experiments were performed on a database of 500 face images. Duplication of data was minimized due to the flight planned in advance. The obtained accuracy of gender recognition is 95.14%, 70% data was used for training.