Publications

Abstract : Emotions frequently aid in the facilitation of human connections. These emotions can be detected in various ways, which include body language, speech, and facial expressions. Many strategies have been used to determine a person's emotion over the years as Artificial Intelligence and Machine Learning have advanced. This paper represents a system that recognizes a person's emotions by looking at their face and adds some quotations to it. A neural network-based method along with image processing is utilized to classify the universal emotions: disgust, scared (fear), sad, surprise, happy and anger. In the proposed model, the “dropout” mechanism which is a regularized method to reduce overfitting and theextended Cohn kanade (CK+) dataset have been used. To improve training efficiency and classification performance, preprocessing and data augmentation approaches are performed. The best structure to capture facial expressions can be obtained from the existing CNN structure by altering convolution layer’s feature map count and making changes corresponding to the fully-connected layer’s nodes. The traditional models were not able to achieve more than 70% accuracy. However, this proposed model achieves 85.7% accuracy with relevant quotations which will be helpful to uplift your mood when you feel low.