Facial emotion recognition and characterization of relationship between health and wellbeing using deep learning | |
| Author | Singh, Vineet Gulab |
| Call Number | AIT Thesis no.RS-23-10 |
| Subject(s) | Neural networks (Computer science) Emotion recognition Well-being--Data processing Deep learning (Machine learning) |
| Note | A thesis submitted in partial fulfillment of the requirements for the degree of Master of Engineering in Remote Sensing and Geographic Information Systems |
| Publisher | Asian Institute of Technology |
| Abstract | Facial Emotional Recognition is an impactful study which is a very important parameter in society's well-being consideration. In this research the Canadian grid node pattem is used to identify the facial landmarks. Grid analysis is the tracking based for the movement of human facial structure, but also it helps with the video frames tracking at the same time. The FAU (Facial Action Units) is considered as the one of most important aspects as in this case total 36 Fau's were identified and overall, seven emotions were captured which were Happy, Sad, Anger, Disgust, Neutral, Surprise, Fear. The optimal hyperparameter of random search are determined the training models and also the discrete values of it, the model us trained in using FER 2013 where the accuracy was obtained as 72.16% and similarly the CK+ dataset which was another dataset used in this case gave the accuracy of 84.25% using the multimodal based fusion CNN. The technique of collecting data entailed obtaining a varied dataset consisting of facial photos that were shot in authentic environments. The dataset in question comprises a diverse array of emotions and incorporates individuals from various age groups, genders, and ethnicities. Furthermore, data pertaining to health and wellbeing, including self-reported health status, lifestyle determinants, and self-assessment of emotional states, was gathered in order to create the definitive basis for understanding the correlation between emotions and overall health. The dataset was subjected to thorough validation procedures in order to assure its high quality and dependability. The process of annotating involves the participation of annotators who were responsible for classifying the emotional states exhibited by the persons seen in the photos.The assessment of inter-rater reliability was conducted. and any discrepancies were resolved through consensus. Furthermore, the information pertaining to health and wellbeing was subjected to cross-validation using surveys and interviews, thereby assuring tha precision and reliability of tha gathered data. The deep learning model is a computational framework that utilizas multiple layers of artificial neural networks to extract and learn hierarchical representations of data.The utilization of a convolutional neural network (CNN) was employed for the purpose of conducting facial emotion recognition on the given dataset. The training process involved utilizing a significant percentage of the dataset, and hyperparameter adjustment was carried out to enhance the model's performance. Several preprocessing strategies, including data augmentation, were employed to improve the model's capacity to generalize across diverse persons and environments. After the establishment of the facial emotion identification model, it was utilized to analyze the dataset with the aim of investigating the correlation between emotions and health and wellness. Statistical analysis, such as correlation studies and regression models, were employed to ascertain potential associations and trends. The findings were additionally verified by taking into consideration demographic and lifestyle parameters to control for potential confounding variables. To assess the precision of the deep learning model, a series of exhaustive accuracy tests were conducted. The evaluation measures employed in these assessments encompassed accuracy, precision, recall, and Fl-score to assess the performance of emotion recognition. Furthermore, the veracity of the depiction of the correlation between emotions and health and wellbeing was evaluated via the process of cross-validation and hypothesis testing. The topic of facial expression recognition has garnered significant interest in recent years, mostly because of its potential applications across a range of disciplines, including the healthcarc sector. The first objective of this research is to design and implement a deep learning model that can effectively recognize facial emotions. The subsequent aim is to utilize this model to investigate the correlation between an individual's emotional state and their overall healthı and wellness. The objective of this study is to offer significant insights into the potential influcnce of an individual's emotional state on their overall health and well-being. |
| Year | 2023 |
| Type | Thesis |
| School | School of Engineering and Technology |
| Department | Department of Information and Communications Technologies (DICT) |
| Academic Program/FoS | Remote Sensing and Geographic Information Systems (RS) |
| Chairperson(s) | Tripathi, Nitin Kumar |
| Examination Committee(s) | Chaklam Silpasuwanchai;Sarawut Ninsawat;Srinivasan, Kathiravan |
| Scholarship Donor(s) | AIT Scholarship |
| Degree | Thesis (M.Eng.) - Asian Institute of Technology, 2023 |