A Novel Framework for Facial Emotion Recognition with Noisy and De Noisy Techniques Applied in Data Pre-Processing

A facial expression is a natural representation of human sentiments. It is in the nature of the human to reciprocate to the living world through the facial expression from which the inputs are interpreted. Human science evaluates emotion, feeling, and sentiment by looking at the human face and its curves, but artificially recognizing emotion with high accuracy and fewer computing resources is more difficult. Deep learning was used to build a state-of-the-art technique that efficiently detects the emotion of seven categories, namely Happy, Anger, Sad, Disgust, Neutral, Surprise, and Fear on FER 2013 dataset which consists of 35,887 images in total. In this work an FERCNN model is proposed that is robust to different environments noisy, denoisy and edge filtered. Along with the model, a hybrid feature selection technique was developed, which results in hybrid image datasets for various environments, which are then given to the proposed FERCNN model for computation. When compared to traditional models like the VGG19, Resnet 50, and Xception emotions are properly recognized with higher accuracy throughout both the training and testing stages by employing a hybrid feature selection approach and the FERCNN model one after the other. For one epoch, 64 images are evaluated, with 80% being used for training and 20% being used for testing. The testing accuracy improves dramatically with each epoch. Testing accuracies of 82%, 79%, and 71% were obtained in denoisy, noisy, and edge filtered hybrid techniques on the proposed model. whereas the test accuracies obtained on the VGG19, Renset 50, and Xception models are 63%, 62%, and 58% in the denoisy case, 56%, 74%, and 57% in the noisy case, and 46%, 53%, and 33% in the edge filtered case.This paper's distinctive contribution is a hybrid model of feature selection and a FERCNN model for emotion identification in one of seven classes in different environments.