A Face Recognition Algorithm Based on Convolutional Neural Network

Abstract

In recent years, face recognition has emerged as a popular technique for human recognition. The convolutional neural network (CNN) is a desirable tool for face recognition, due to its accuracy in feature extraction and simplicity in recognition. However, the traditional CNN has several defects, such as high computing load and lack of universality. To overcome these defects, this paper puts forward a novel CNN algorithm for face recognition. Unlike the traditional CNN with a fixed structure, our CNN automatically expands the number of initial neurons based on the pre-set training errors. In addition, the branch structure, network expansion interval and spreading factor were all optimized in the proposed CNN. To verify its performance, the proposed CNN algorithm for face recognition was compared with the traditional CNN through experiments on the ORL face database. The results show that our CNN algorithm outperformed the traditional algorithm in recognition accuracy, and strikes a perfect balance between training time and test error. The research results shed important new light on the application of deep learning in face recognition