Facial Emotional Recognition Using Convolutional Neural Network

Abstract

The face of a person is his identity. Most of his emotions, issues can be derived from his face. Face is the window to the soul which was said by famous French doctor Duchenne de Boulogne. He used several techniques to support his theory like giving shocks of electric impulses to understand how a person reacts to muscular contractions. He also tried to induce some of the expressions of bizarre looking. The ultimate aim is to analyze how much muscles contribute to emotion. He successfully derived lots of human emotions which is hidden. After 200 years this field is still active as it requires more experiments to extract invisible truths of human truths. Lots of Emotion Recognition which is automatic have been seen in the field of Marketing and Advertising. It is also seen in the field of Medical, Law and order$\ldots$etc. The main fundamental question arrives whether it is good to enter into our personal space. This a question which has lots of dimension. Those who are against this practice is claiming that this is a violation of human rights by which we can use the stored data to harm the society in future. Even though it is a major concern if we can rectify the above the scope in this area is phenomenal. Lots of researchers are still working on this area because of mainly one reason that is its wide application in the area of Medical Science. It can be used by a doctor for diagnosing a person with certain psychological disorders, children with Autism, person with Parkinson’s disease. It can be widely used to diagnose children with Autism so that particular child can be motivated at early stage leads to his success in future and thus provide a good citizen to the society. In our research we are using several datasets such as FER2013, CK+$\ldots$etc. These datasets have been given to a model which includes Deep Convolutional Neural Network. Here input image is given to a model using camera. The particular image has to done preprocessing to extract fine features in the model. After pre-processing we should extract emotions from the photo which is given as input by comparing it with datasets and extract the emotions from the given photo. By using FER2013 dataset the author got validation accuracy of around 99.3347 percentage and test accuracy around 60.94 percentage, with CK + it is around 96.87 percentage and 71.62 percentage for validation and test respectively.