Facial Expressions Recognition Using Markov Stationary Feature - Vector Quantization And Support Vector Machine Method
Irfan Maliki, Fascal Sapty Jarockohir
UNIKOM Bandung
Abstract
Facial Expression is one form of communication with humans. Every human being must have different forms of facial expressions, this causes difficulties for the computer to recognize it. In the process of introducing facial machine learning expressions do not recognize the expression directly, but the feature extraction process needs to be done first. One feature extraction method that can be used is Markov Stationary Feature - Vector Quantization (MSF-VQ). The MSF-VQ algorithm has been tested in the case of human face recognition and has an accuracy of 99.16%. In this study the MSF-VQ algorithm is used to recognize facial expressions. The total data used for training data and testing data was 1440. For training data 1170 facial photos were obtained from 15 people, with 6 types of facial expressions, and from one type of expression there were 13 facial photos per person. The test was carried out using 270 test data using the MSF-VQ algorithm using a Multiclass Support Vector Machine learning machine with a linear kernel. Based on the results of tests that have been conducted, the level of accuracy achieved using the MSF-VQ method reaches 97.14%. This shows that the MSF-VQ method can be recommended as a good method for facial expressions recognition.
Keywords: Facial Expressions recognition, MSF-VQ method, SVM method
Topic: Informatic and Information System