A NOVEL DEEP LEARNING BASED ARCHITECTURE FOR FACIAL GESTURE ANALYSIS

Emek Soylu B., Güzel M. S., Askerbeyli İ.

APPLIED AND COMPUTATIONAL MATHEMATICS, cilt.19, sa.3, ss.300-316, 2020 (SCI-Expanded)

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 19 Sayı: 3
  • Basım Tarihi: 2020
  • Dergi Adı: APPLIED AND COMPUTATIONAL MATHEMATICS
  • Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Aerospace Database, Communication Abstracts, Metadex, zbMATH, Civil Engineering Abstracts
  • Sayfa Sayıları: ss.300-316
  • Ankara Üniversitesi Adresli: Evet

Özet

Facial gestures carry critical information as non-verbal communication and are considered one of the critical problems of computer vision.  Deep learning is emerging as a powerful approach for machine learning and has achieved satisfactory performance in many areas. This paper proposes a novel facial gestures analysis system, implementing a novel deep neural network structure based on a Convolutional Neural Network (CNN) architecture. Seven different gesture classes are defined for the facial gesture analysis. Various facial images are obtained from different comprehensive datasets to train and verify the overall performance of the proposed system. Having observed the performance of the proposed system on the facial gesture analysis problem, it is first compared with a conventional system currently based on objective evaluation parameters, which is then compared with recent studies that rely on deep learning for the same problem.  Experimental results, based on JAFFE, KDEF, and MUG databases, verify the superiority of the proposed system over the conventional approach and also deep learning based systems.


Keywords: Facial Gesture Analysis, Deep Learning, CNN, HOG, SVM.