Hybrid Model for Emotion Recognition Based On Multiple Features in Video Sequences

Abstract

Facial emotion recognition in video is one of the efficient task which is used to convey the
expressions of human beings. Due to various issues like age, different facial features of one person to
other so that an automatic video facial emotion analysis and recognition is a challenging task. In
order to overcome this challenges, an intelligent and novel emotion recognition model is proposed in
this work. Our method employed guided image filtering technique for filtering also incorporated
feature extraction techniques as geometric features and Local Binary Pattern (LBP). After that, the
Kernel Entropy Component Analysis (KECA) is used for dimension reduction. Finally, this work
employed a Deep Recurrent Neural Network (DRNN) as the classifier for the emotion identification
and here complexities of the weight updating is reduced by introducing the Adaptive Galactic Swarm
Optimization (AGSO) algorithm Here, AGSO algorithm is used to optimize parameters and initial
connection weights that control the efficiency of the DRNN. The proposed work is performed in the
MATLAB environment. Experiments performed on FAMED, EMMA and RAVDESS datasets and it
provides a classification accuracy of 98.33%, 98.02% and 97.96% respectively, this is higher than
other current procedures for six-class emotion.