The Neural Network Technologies Effectiveness for Face Detection

The paper is devoted to estimation of the effectiveness of the use of modern convolutional neural networks for face detection. On standard open datasets, learning of neural networks and comparison of the effectiveness of their functioning are carried out. Conclusions are drawn regarding the practical application of the neural networks for detecting faces on digital photographs.

I. Introduction

The problem of face recognition has become very important in the field of biometric identification. Facial detection in digital images has many applications in solving real-life problems, including entertainment, healthcare, security, etc. [1]. To this day, there are a lot of approaches and models for solving these tasks. Although older algorithms, such as the Viola-Jones algorithm perform facial recognition tasks fast and accurately in close to optimal conditions, their performance diminishes when input images are distorted, too dark or too bright, or when faces appear in the image under different angles [2]. The development of deep learning and deep convolutional neural networks (NN) allowed researches to invent complex neural network-based machine learning algorithms which, when properly trained, perform much better and reliably on images with faces in different environments, angles and lightning conditions [2]. Also, each of the emerged neural network-based algorithms has its features and characteristics which determine its efficiency when applied to different types of applications.