Hybrid feature extraction method based-on probability model for augmented facial expression recognition
Thesis (PhD. (Computer Science))
Main Author: | |
---|---|
Format: | Thesis |
Language: | English |
Published: |
Universiti Teknologi Malaysia
2024
|
Subjects: | |
Online Access: | http://openscience.utm.my/handle/123456789/969 |
_version_ | 1796848940415451136 |
---|---|
author | Jenni, Kommineni |
author_facet | Jenni, Kommineni |
author_sort | Jenni, Kommineni |
collection | OpenScience |
description | Thesis (PhD. (Computer Science)) |
first_indexed | 2024-03-05T17:35:12Z |
format | Thesis |
id | oai:openscience.utm.my:123456789/969 |
institution | Universiti Teknologi Malaysia - OpenScience |
language | English |
last_indexed | 2024-03-05T17:35:12Z |
publishDate | 2024 |
publisher | Universiti Teknologi Malaysia |
record_format | dspace |
spelling | oai:openscience.utm.my:123456789/9692024-01-16T12:00:36Z Hybrid feature extraction method based-on probability model for augmented facial expression recognition Jenni, Kommineni Human-computer interaction—Research Face perception—Computer simulation Emotions—Computer simulation Thesis (PhD. (Computer Science)) In human communication, facial expressions play a significant role and offer rich information about human intentions, emotions such as anger, disgust, fear, happy, neutral, sad, and surprise due to the outer world circumstances or inner feelings. Facial Expression Recognition (FER) acts as an essential method for understanding human emotional behaviour. Over the last two decades, automatic FER has acted as one of the multimedia challenging research areas happening in human-computer interaction, facial, and pattern recognition. There are several automatic systems with precise recognition rate that have been proposed to recognize facial emotions in an image. However, challenges still exist because of the dynamic nature of the human face based on outer world circumstances or inner emotions. Researchers are struggling hard to identify suitable feature extraction procedures from images due to the fact of their uncertain characteristics of emotion expression. The research observes that there is no suitable comprehensive study on emotion detection and FER algorithms. Besides, the existing feature extraction algorithms are not accurate for emotion recognition and classification, in which it leads to weakness in its accuracy measure. Moreover, there is no appropriate evolution that has been developed for the hybrid algorithm. Thus, this research focuses on the following objectives: to develop taxonomy for facial expression recognition algorithms; and then to propose a new hybrid of Dual-Tree Multi-Band Wavelet Transform (DTMBWT) and Grey Level Cooccurrence Matrix (GLCM) based feature extraction algorithm in getting precise facial information from the human face input images to achieve high accuracy. Subsequently, the final objective is to evaluate the performance of the proposed algorithm using accuracy metrics for facial expression classification. To achieve these objectives, this research develops taxonomy of emotion detection and facial expression classification algorithms for problem formulation in which it is then followed by studying the effect of a single algorithm for feature extraction using DTMBWT with energy and entropy. The energy and entropy features are computed for each sub-band of DTMBWT of each class of emotion and are utilized as features for corresponding facial image. In addition, to increase the classification accuracy, this research proposed a new hybrid algorithm with DTMBWT and GLCM based contrast and homogeneity. In the final stage, by combining all extracted feature vectors, the proposed method employs a Gaussian Mixture Model (GMM) as a classifier for the classification. GMM is an efficient density estimator in which a probability density function is modelled by a summation of Gaussian distributions. The experiment is carried out using Matlab and Japanese Female Facial Expression (JAFFE) database. The precision in terms of accuracy of the proposed hybrid system is recognized with qualified observations and analysis using various levels of DTMBWT decomposition and several Gaussians used in the GMM classifier. The experiments result (98.14%) in precision is recorded using a single feature extraction algorithm which is DTMBWT. Subsequently the experimental results (99.53%) are observed for the hybrid method which is comparatively high and significantly enhanced with the utmost accuracy of the proposed technique. Looking at the experiment results, it can be concluded that the proposed hybrid method gives an enhanced accuracy. The significance of the proposed research contributes in enhancing knowledge in computer vision, its computing methods of FER and is beneficial to surveillance systems, healthcare industry, researchers, and psychological studies. Faculty of Engineering - School of Computing 2024-01-16T03:46:55Z 2024-01-16T03:46:55Z 2021 Thesis Dataset http://openscience.utm.my/handle/123456789/969 en application/pdf application/pdf application/pdf application/pdf application/pdf Universiti Teknologi Malaysia |
spellingShingle | Human-computer interaction—Research Face perception—Computer simulation Emotions—Computer simulation Jenni, Kommineni Hybrid feature extraction method based-on probability model for augmented facial expression recognition |
title | Hybrid feature extraction method based-on probability model for augmented facial expression recognition |
title_full | Hybrid feature extraction method based-on probability model for augmented facial expression recognition |
title_fullStr | Hybrid feature extraction method based-on probability model for augmented facial expression recognition |
title_full_unstemmed | Hybrid feature extraction method based-on probability model for augmented facial expression recognition |
title_short | Hybrid feature extraction method based-on probability model for augmented facial expression recognition |
title_sort | hybrid feature extraction method based on probability model for augmented facial expression recognition |
topic | Human-computer interaction—Research Face perception—Computer simulation Emotions—Computer simulation |
url | http://openscience.utm.my/handle/123456789/969 |
work_keys_str_mv | AT jennikommineni hybridfeatureextractionmethodbasedonprobabilitymodelforaugmentedfacialexpressionrecognition |