New Geometric Based Features for Facial Expression Recognition

EasyChair Preprint 4864, version 3

Abstract

Facial expression recognition has significant benefit due to its possible applications in computer vision. Facial expression recognition include machine and human interaction, security, psychology, and changing appearance in social media applications. In this paper, new geometric based features are proposed for the recognition of seven facial expressions. The number of features is reduced by using feature selection methods. Obtained features are applied to Support Vector Machines (SVM) classifier. In the experimental studies, the extended Cohn-Kanade (CK+) dataset is used and facial expressions was classified by the 10-fold cross-validation method. Classification accuracy of 93.5% was achieved in CK+ data with the features selected by the Sequential Backward Feature Selection method.

Keyphrases: Support Vector Machine, facial expression recognition, feature selection, geometric based features

@booklet{EasyChair:4864,
  author    = {Nuri Özbey and Mehmet Bilginer Gülmezoğlu},
  title     = {New Geometric Based Features for Facial Expression Recognition},
  howpublished = {EasyChair Preprint 4864},
  year      = {EasyChair, 2021}}