Real-time Face Recognition using SVM, MLP and CNN

Abstract

Real-time face recognition targets to match a human’s face in a digital image or a video frame against faces in a database, which plays an important role in various fields nowadays. Current face recognition methods use different classification algorithms to achieve their functionalities for different purposes such as identification and entertainment. However, most current face recognition methods are fulfilled by only one algorithm, which causes inaccurate predictions. This paper proposes a method that fuses support vector machine (SVM), Multi-layer Perceptron (MLP) and Convolutional Neural Network (CNN), into a single real-time face recognition system for Real-time face recognition and enhancing the prediction accuracy. Specifically, we first crawl the data from the internet, process them into a 1-D array and store them into the database for the first step. Then, the data from the database is used to train models including SVM, MLP and CNN. Next, we fused these three algorithms to develop the application and calibrate the camera to reduce distortion. Finally, we develop the front-end interface of this application using the Tkinter module. To verify the effectiveness of our method, we compare our methods with SVM, MLP and CNN. The accuracy of the fused algorithm is 90%. In comparison, the accuracies of SVM, MLP and CNN are 60%, 70% and 80%, respectively. The results demonstrate that the fused algorithm outperforms the other three situations and have the potential to contribute to the future development of real-time face recognition systems.