A Novel Lightweight Deep Convolutional Neural Network Model for Human Emotions Recognition in Diverse Environments

A picture is worth a thousand words to convey information during daily life communication. Recognizing person’s emotions from facial expressions (FEs) has become prevalent research field in the past few decades. Deep learning (DL) models, mainly deep convolutional neural networks (DCNNs), are trending last years perform recognition/classification tasks. Several prior DCNNs have resulted good recognition accuracy for emotion (FER) systems. Still, there need an effective as well efficient FER system that can recognize FEs irrespective of illumination conditions, subjects’ gender, age range, geographical locations, race, etc. In this work, we presented novel lightweight DCNN model aforementioned conditions. We made it by optimal selection its hidden layers, which ultimately reduced number floating point operations (FLOPs). embedded data augmentation step into model’s training phase enhance generalization ability. An early stopping criterion also introduced prevent overfitting. trained and evaluated performance proposed on widely used benchmarks databases. selected five diverse databases (two collected lab, one based stylized cartoon characters, two collections unconstrained realistic environment), including CK+, Karolinska Directed Emotional Faces (KDEF), Facial Expression Research Group (FERG), Recognition-2013 (FER-2013), Real-World Affective Database (RaF-DB). The 99.98%, 99.25%, 88.17%, 84.09%, 69.87%, 69.16% achieved FERG, KDEF, RaF-DB, FER-2013 human faces (FER-2013H), complete (FER-2013C) database, respectively. Our outperforms state-of-the-art approaches terms FLOPs.