Malaysia Ethnicity-based Facial Expression Classification and Emotion Mapping

• Main Author: Izzah Nilamsyukriyah, Buang
• Format: Thesis
• Language: English
• Published: Universiti Malaysia Sarawak (UNIMAS) 2021
• Subjects: QA75 Electronic computers. Computer science
• Online Access: http://ir.unimas.my/id/eprint/35187/3/Izzah.pdf; http://ir.unimas.my/id/eprint/35187/

Human are commonly good in notifying several emotions via facial expression. In daily human communication, it is crucial for each person to be able to convey his emotions and perceive others respectively using speech, facial expressions and body movements. The computer vision experts are continuously learning on how to achieve high performance in analysing faces, especially which occur spontaneously. Malaysian facial database and analysis are still inconspicuous, especially for local ethnicity studies. Hence, this thesis developed MUA Database, the first Malaysian ethnicity facial database, which consists of data from non-actor subjects from 4 local ethnicities that are Chinese, Iban, Indian and Malay. During the data collection, the subjects are encouraged to express facial expressions spontaneously. Facial expressions analyses are done using the database and facial deformation for each ethnicity is evaluated. From the experiments, the performance of HOG, LBP and SIFT are compared for feature extraction, and SVM, Decision Tree and KNN performance are evaluated as classifier. Results show that the combination of HOG features and KNN classifiers are the best pair for ethnic recognition with 96.90% accuracy, whereas HOG features and SVM classifier combination shows the best pair for emotion recognition with 59.10% accuracy. Indian appeared to be the most recognisable among other ethnicities. As for emotion, “happy” appear to be the most conspicuous emotion, whereas “fear” is the least visible among all tested emotion.