Title: Facial expression recognition using local multidirectional score pattern descriptor and modified hidden Markov model

Authors: Mayur Rahul; Narendra Kohli; Rashi Agarwal

Addresses: Department of Computer Applications, University Institute of Engineering and Technology, Chhatrapati Shahu Ji Maharaj University, Kanpur, India ' Department of Computer Science and Engineering, Harcourt Butler Technical University, Kanpur, India ' Department of Information Technology, University Institute of Engineering and Technology, Chhatrapati Shahu Ji Maharaj University, Kanpur, India

Abstract: Facial expression recognition is an application used for biometric software that can be used to recognise special expressions in a digital image by comparing and analysing different patterns. Recognising faces becomes very difficult when there is a frequent change in facial expressions. In this paper, two tier modification of normal hidden Markov model (HMM) is used to identify continuous effective facial expressions. Two tier extension of HMM: bottom and upper tier. Bottom tier represents the atomic facial expressions which are made by mouth, eyes and nose separately and upper tier represents the joint of these atomic facial expressions such as neutral, sadness etc. This paper introduces the improved modified local decision based unsymmetrical trimmed median filter (IMLDBUTMF) for noise reduction, the local multidirectional score pattern (LMSP) for feature extraction and modified hidden Markov model for classification process. The experimental result shows that the proposed method gives enhanced performance of 85% of recognition rate on publicly available datasets JAFFE.

Keywords: facial expression recognition; feature extraction; machine learning; hidden Markov model; HMM; Baum Welch method; forward method.

DOI: 10.1504/IJAIP.2021.113787

International Journal of Advanced Intelligence Paradigms, 2021 Vol.18 No.4, pp.538 - 551

Received: 06 Aug 2018
Accepted: 09 Sep 2018

Published online: 31 Mar 2021 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article