Abstract:

Abstract Face detection and emotion recognition are two closely connected tasks in computer vision that include analysing facial images to identify faces and detect the emotions expressed by the individual. Face detection is the way of localizing and locating faces within image or video frames. The objective is to detect the presence and position of faces, by drawing bounding boxes around them. Facial emotion recognition (FER) aims to detect and classify the emotions expressed by individuals based on facial expressions. Typically, this task can be done after face detection, where the faces detected are analysed further for emotional cues. Emotion recognition can be advanced by means of classical deep learning (DL) or machine learning (ML) techniques. Contemporary research on emotion classification has accomplished grand performance over DL based approaches. This article introduces an Optimal Deep Convolutional Neural Network based Face Detection and Emotion Recognition model (ODCNN-FDER) technique. The aim of the ODCNN-FDER technique is to detect faces and identify the existence of different emotions in them. To achieve this, the ODCNN-FDER technique initially employs Multi-Task Cascaded Convolutional Neural Network (MCCNN) model. Next, the fusion based feature extraction process is involved using two DL models namely EfficientNetB3 and InceptionResNetV2. For emotion recognition, Convolutional Attention Gated Recurrent Neural Network (CAGRNN) model is used. Lastly, root mean square propagation (RMSProp) optimizer was exploited for the optimal hyperparameter tuning of the CAGRNN approach. The performance validation of the ODCNN-FDER methodology was tested on the FER-2013 database. The experimental values highlighted the improved face detection and FER results of the ODCNN-FDER technique over other models.

Keywords: