A Hybrid Approach based on Haar Cascade, Softmax, and CNN for Human Face Recognition

A HYBRID APPROACH FOR HUMAN FACE RECOGNITION

Abstract

Face recognition has been studied long but it is still an important and current research field in deep learning, computer vision, and forensics. There are several applications such as group action systems, human-machine interaction, and security systems, where face recognition is of vital importance. It is noticed that the algorithms based on Deep Learning (DL) have shown higher performances, stipulation of accuracy, and processing speed as compared to traditional machine learning algorithms. With its dominant methodology in deep learning, the Convolutional Neural Network (CNN) has contributed immensely to face recognition. In this paper, a novel hybrid version of the deep learning algorithm containing Haar Cascade, SoftMax, and CNN components is proposed. It provides promising results for applications based on the recognition of human faces. In the experiments, the accuracy of this hybrid algorithm is achieved at 99.95%, which is significantly higher than existing Viola-Jonas and Principal Component Analysis (PCA), which have accuracy rates of 74.38% and 81.81% respectively. However, the accuracy of our proposed algorithm close to Linear Discriminant Analysis (LDA) at 95.45%, and SoftMax and CNN at 94%. In this paper, the proposed hybrid deep learning algorithm improves the result performance and is compared with some existing techniques for face recognition.