Abstract:

Abstract Human emotions can be identified by recognizing their facial expressions. This relates to the applications such as medical diagnostics, human emotional analysis, human-robot interaction, etc. This study presents a novel Convolution Neural Network (CNN) model for recognizing emotions from facial images. The proposed model, “ConvNet-3”, recognizes the emotions such as disgust, anger, fear, happy, sad, surprise and neutral. The main focus of the proposed research is on training accuracy of the model in lesser number of epochs. The proposed model is trained on FER2013 dataset and its performance is evaluated. ConvNet-3 consists of 3 convolution layers and two fully connected layers. As illustrated in experimental results, the ConvNet-3 obtains training accuracy of 88% and validation accuracy of 61% on FER2013 which is better than existing models. In contrast it is observed that the presented model over fits on CK+48 dataset. 

Keywords: