Face Detection and Recognition From Distance Based on Deep Learning

Introduction

With the rapid development of computer vision, artificial intelligence (AI) has become the core of contemporary high-tech, its applications include face detection and recognition. The key issue in face detection and recognition is feature extraction which has been developed rapidly based on computational intelligence (Bonetto et al. 2015; Wang and Srinivasan, 2017). There are many types of applications of face recognition, e.g., gender identification, ages and emotions are the most important characteristics of human faces. The main purpose of this book chapter is to automatically recognize human faces and affirm the class of the objects. This topic is also one of the important research fields in computer vision and deep learning (LeCun et al., 2015).

Face recognition, as a biometric, is a significant components of video surveillance and visual security which has been applied to human identification nowadays. In large shopping malls, face recognition is employed to monitor the passengers and provide users with convenient services. At the entrance of a railway station, airport, bank, supermarket, school or company, face recognition is applied to implement access control, which prevents the entry of aliens and ensures the security of premise (Zhang, et al., 2017). With the development of human face recognition, it has been applied to protect the privacy of users, improve the security of visual data. Face recognition, as a special part of human-computer interaction (Bian et al., 2016) identifies users, serves them with great convenience.

In 1943, McCulloch and Walter (Landahl et al., 1943) proposed that the first artificial neuron model: MP model who connected the basic unit together to understand how human brain produces highly complex patterns. This has made a significant contribution to the development of artificial neural networks. In 1958, Rosenblatt greatly developed the neural network theory and applied it to real problems. In 1986, Rinehart et al. proposed backpropagation algorithm, which is an important method in the neural network to calculate the errors of neurons after data processing (Liu, and Liang, 2005). This algorithm is still the most popular one of the most broadly applied artificial neural networks in artificial intelligence.

With the development of neuroscience, computer science scientists have found that brain signals are transmitted through a complex structure; if time permits, the characteristics are applied to understand digital signals, which led to the emergence of deep learning (LeCun et al., 2015) for the establishment and simulation of human brain for analysis and learning. Convolutional neural networks (CNNs) (Karpathy et al., 2014) have been successfully applied to visual imagery in the past few years. One of the most important factors is the need to provide a large amount of training data. But in face recognition, due to lack of large scale of data sets, a few of experiments were limited.

The contribution of this book chapter is based on deep learning for face recognition, which will be completed in real time. For example, if people are near to the camera, the system will verify the influence of proportion of face to the total size of the given images, the proportion is thought as a major core. The relevant experiments require four parts: 1) Collecting the data set, 2) accepting the command parameters, and 3) defining the neural network model, 4) training and testing model.

This book chapter is organized as follow. Literature review is presented after this introduction section, then the method and final results are demonstrated, the conclusion is drawn at last.