Facial Expression Recognition Using Wavelet Transform and Convolutional Neural Network

Abstract

A facial expression recognition is one of the machine learning applications. It categorizes an image of facial expression into one of the facial expression classes based on the extracted features from an image. Convolutional Neural Network (CNN) is one of the classification methods in which also extracts patterns from an image. In this research, we applied the CNN method to recognize facial expression. The wavelet transform is used before being processed into CNN to improve the accuracy of facial expression recognition. The facial expression images are taken from Karolinska Directed Emotional Faces (KDEF) dataset which contains seven different facial expressions. The preprocessing of the images includes converting the image to grayscale, changing the image resolution to 256 × 256 pixels, and applying data augmentation with horizontal reflection and zoom in. The experimental results of facial expression recognition using CNN with wavelet transform achieve 84.68% accuracy and without wavelet transform achieve 81.6%. The best result is 89.6% accuracy which is obtained with the data split based on the photo session, using wavelet transform, RMSprop optimizer with learning rate 0.001, and without data augmentation.