Facial Expression Recognition Using Extreme Learning Machine

Facial expression recognition is one of the technological capabilities in identifying a face image to follow up on research conducted by psychologists. The recognition of facial expressions is very important to know the emotions of someone who is experiencing it. In this study two datasets were used, namely the FER2013 and CK + datasets. The FER2013 dataset and CK+ are datasets designed to identify facial expressions. At the feature extraction stage, it uses the Histogram of Oriented Gradient (HOG) feature dan Local Binary Pattern (LBP) feature. Whereas in the classification stage, the Extreme Learning Machine (ELM) classifier is used. The greatest accuracy by using HOG feature is 63.86% for the FER2013 dataset and 99.79% for the CK + dataset with sigmoid as an activation function. And the greatest accuracy by using LBP feature is 55.11% for the FER2013 dataset and 98.72% for the CK + dataset with RBF as an activation function.