Facial Expression Recognition for the Blind Using Deep Learning

A large number of people living around us are visually impaired. One of the most difficult tasks faced by them is the identification of the expression of the people in front of them. They are not aware of the intentions and emotions of the other person. Thus a system to assist the blind in recognizing the facial expressions of the confronting person can be of great use. A facial expression recognition system was developed using the convolutional neural network methodology in deep learning. Two models were created for the same. The first model was a proposed CNN architecture trained using FER-2013 dataset. The model could classify the expressions into 7 different classes and obtained an accuracy of 67.18%. The second model was based on transfer learning approach trained using cleansed FER-2013 dataset. The model could classify the expressions into 4 different classes and obtained an accuracy of 75.55%. The optimized model of the original transfer learning model was created and deployed on the android device. The model can capture the image of the other person and provide the corresponding class label of expression. The classified text will be then converted to speech for assisting the blind.