Open AccessMulti-Pose Facial Expression Recognition Using Hybrid Deep Learning Model with Improved Variant of Gravitational Search Algorithm
Author(s) -
Yogesh Kumar,
Shashi Kant Verma,
Sandeep Sharma
Publication year - 2022
Publication title -
the international arab journal of information technology
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.227
H-Index - 27
eISSN - 2309-4524
pISSN - 1683-3198
DOI - 10.34028/iajit/19/2/15
Subject(s) - computer science , artificial intelligence , convolutional neural network , deep learning , discriminative model , facial expression , pattern recognition (psychology) , feature (linguistics) , local binary patterns , deep belief network , algorithm , machine learning , histogram , image (mathematics) , philosophy , linguistics
The recognition of human facial expressions with the variation of poses is one of the challenging tasks in real-time applications such as human physiological interaction detection, intention analysis, marketing interest evaluation, mental disease diagnosis, etc. This research work addresses the problem of expression recognition from different facial poses at the yaw angle. The major contribution of the paper is the proposal of an autonomous pose variant facial expression recognition framework using the amalgamation of a hybrid deep learning model with an improved quantum inspired gravitational search algorithm. The hybrid deep learning model is the integration of the convolutional neural network and recurrent neural network. The applicability of the hybrid deep learning model can be considered as significant if the feature set is efficiently optimized to have the discriminative features respective to each expression class. Here, the Improved Quantum Inspired Gravitational Search Algorithm (IQI-GSA) is utilized for the selection and optimization of features. The IQI-GSA method is significant for optimizing the features compared to quantum-behaved binary gravitation search algorithm for handing the local optima and stochastic characteristics. Comparing with state-of-art techniques, the proposed framework exhibits the outperformed recognition rate for experimentation on Karolinska Directed Emotional Faces (KDEF) and Japanese Female Facial Expression (JAFFE) datasets.