Performance Dependency of Facial Emotion Recognition System on Dropout and Learning Rate

This paper proposes facial expression recognition system developed through machine learning which is capable of recognizing the emotions in real time. The work is divided mainly into two parts. In the first part the face in the image is recognized using Haar cascade classifier and AdaBoost algorithm, the idea introduced by Viola and Jones. This method can be adapted to identify the facial features rapidly with high accuracy rate. In real time application the model is capable of computing 15 frames per second. The second part deals with recognizing the emotions based on facial features identified in the first part. The system classifies number of faces simultaneously in real time. This model is implemented with deep learning algorithms using Convolutional Neural Network. Neural network is trained using the dataset FER2013 to identify different emotions. The task is to attribute emotional states such as happy, sad etc. to subjects based on their facial characteristics. Extracting the emotions from one's profile facilitate the machines to interact with the public in various ways aiding in the development of new innovative applications. Machine learning and deep neural network acquire the information from complex appearance and categorize the patterns.