A COMPARISON OF DEEP LEARNING BASED ARCHITECTURE WITH A CONVENTIONAL APPROACH FOR FACE RECOGNITION PROBLEM

Abstract

This paper addresses a new approach for face recognition problem based on deep learning strategy. In order to verify the performance of the proposed approach, it is compared with a conventional face recognition method by using various comprehensive datasets. The conventional approach employs Histogram of Gradient (HOG) algorithm to extract features and utilizes a multi-class Support Vector Machine (SVM) classifier to train and learn the classification. On the other hand, the proposed deep learning based approaches employ a Convolutional Neural Network (CNN) based architecture and also offer both a SVM and Softmax classifiers respectively for the classification phase. Results reveal that the proposed deep learning architecture using Softmax classifier outperform conventional method by a substantial margin. As well as, the deep learning architecture using Softmax classifier also outperform SVM in almost all cases.