Facial expression serves as an inherently potent and universally understood means for humans to convey their emotions and intentions. The technology for classifying facial expressions has numerous practical applications across different fields, including emotion awareness in humanoid robots. Although there has been notable progress in deep learning methods for facial expression classification (FEC), a persisting challenge lies for FEC to develop an embedded system compatible method with a limited size of facial expression dataset. While keeping in mind the constraint of small dataset, it is worth studying the potential of available state-of-the-art (SOTA) methods for the application FEC. We have implemented three strategies (i.e., feature extraction, partially freezing, and full fine-tuning) on five (VGG16, ResNet50, MobileNetV2, EfficientNetB2, and DenseNet121) different SOTA Deep Convolutional Neural Network (DCNN) architectures. Feature extraction strategy involves training fully connected layers only while the partially freezing and full fine-tuning involve fine-tuning with unfreezing last block and full convolutional base of pre-trained DCNN model. In total, a detailed comparison is carried out on 15 DCNN models using Extended Cohn-Kanade (CK + \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$+$$\end{document}) dataset, which is laboratory-controlled dataset. Results demonstrate that the full fine-tuning strategy on MobileNetV2 outperforms the other 14 architectures with 94.74% test set (unseen data) accuracy. While VGG16 performed best in partially freezing and feature extraction with test set accuracy 87.37% and 81.05% respectively. Further this study indicates that the proposed three transfer learning-based strategies can play pivotal role to obtain optimal DCNN model for facial expressions classification.


    Access

    Check access

    Check availability in my library

    Order at Subito €


    Export, share and cite



    Title :

    Emotion Awareness in Humanoids: Human Facial Expressions Recognition Using Transferred Deep CNN Models


    Contributors:

    Published in:

    Publication date :

    2024-07-24


    Size :

    18 pages




    Type of media :

    Article/Chapter (Book)


    Type of material :

    Electronic Resource


    Language :

    English




    Automatic Human Emotion Recognition System using Facial Expressions with Convolution Neural Network

    MADUPU, RAM KUMAR / KOTHAPALLI, CHIRANJEEVI / YARRA, VASANTHI et al. | IEEE | 2020



    Automatic Recognition of Human Facial Expressions

    Matsuno, K. / Lee, C.-W. / Kimura, S. et al. | British Library Conference Proceedings | 1995


    Automatic recognition of human facial expressions

    Matsuno, K. / Chil-Woo Lee / Kimura, S. et al. | IEEE | 1995


    CNN based Recognition of Emotion and Speech from Gestures and Facial Expressions

    Avula, Himaja / R, Ranjith / S Pillai, Anju | IEEE | 2022