Multi-level graded facial emotion intensity recognition using MCANN for health care

Abstract

Facial emotion recognition analysis is widely used in various social fields, including Law Enforcement for police interrogation, virtual assistants, hospitals for understanding patients’ expressions, etc. In the field of medical treatment such as psychologically affected patients, patients undergoing difficult surgeries, etc require emotional recognition in real-time. The current emotional analysis employs interest points as landmarks in facial images affected by a few emotions Many researchers have proposed 7 different types of emotions (amusement, anger, disgust, fear, and sadness). In our work, we propose a deep learning-based multi-level graded facial emotions of 21 different types with our proposed facial emotional feature extraction technique called as Deep Facial Action Extraction Units (DFAEU). Then using our Multi-Class Artificial Neural Network (MCANN) architecture the model is trained to classify different emotions. The proposed method makes use of VGG-16 for the analysis of emotion grades. The performance of our model is evaluated using two algorithms Sparse Batch Normalization CNN (SBN-CNN) and CNN with Attention mechanism (ACNN) along with datasets Facial Emotion Recognition Challenge (FERC-2013). Our model outperforms 86.34 percent and 98.6 percent precision.