文章基本信息

标题：BENCHMARKING OF CONVOLUTIONAL NEURAL NETWORKS FOR FACIAL EXPRESSIONS RECOGNITION
本地全文：下载
作者：AHMED HESHAM MOSTAFA ; HALA ABDEL-GALIL EL-SAYED ; MOHAMED ABDEL-FATTAH BELAL 等
期刊名称：Journal of Theoretical and Applied Information Technology
印刷版ISSN：1992-8645
电子版ISSN：1817-3195
出版年度：2020
卷号：98
期号：18
页码：3104-3115
出版社：Journal of Theoretical and Applied
摘要：Facial Expression Recognition (FER) is one of the most interesting problems in computer science due to its potential applications in AI, many studies were proposed for the FER, but it based on traditional machine learning techniques and these techniques do not have the generalizability to classify expressions from unseen images or those that are captured from the wild, so it is still a diﬃcult and a complex problem. Recently, trends of research in various ﬁelds have begun to transfer to deep learning techniques, since it can learn and capture features automatically, robustness to natural variations in the data and generalizability. This work presents a comparative analysis of the popular convolutional neural network (CNN) models based on modular CNN architectures such as ResNet, DenseNet, MobileNet, NASNetMobile, Inception and Xception, applied on FER problem. The purpose of the paper is benchmarking the best architecture models, in order to help researchers to explore and investigate the best architectures for future research in FER based CNN models. For the comparative analysis multiple metrics were used such as Accuracy, Loss, precision, recall, number of parameters and model size, to conduct experiments facial expressions dataset was used from the AffectNet dataset with 287,651 images for training and 4000 images for validation represent eight facial expressions.
关键词：Convolutional Neural Network;Deep learning;Emotion Recognition;FER;Facial Expression Recognition