Lightweight Attention Based Deep CNN Framework for Human Facial Emotion Detection from Video Sequences

Emotions with their intensities are associated with the action of humans which decides the behaviour of an individual.The recent research has gained enormous attention in the domain of emotion detection due to automatic facial emotion detection. The prime goal of facial emotion recognition (FER) is intended to map different facial expressions from video sequences to specific emotional states. Human facial emotion recognition (HFER) has captivated the attention of researchers for its excellent real time application which serves the society. This research work has proposed a novel lightweight attention based deep convolution neural framework for human facial emotion detection from video sequences on CK + dataset. A dense convolutional network (DenseNet-201) has been applied to strengthen feature propagation to find a solution to the vanishing-gradient problem, and significantly lower the parameter count.To provide the model for better visual perceptibility, an attention block is applied between the convolution and dense layer. The proposed model achieves good accuracy of 98.8% while using less memory and processing in comparison to state-of-the-art approaches to reach good performance on CK + data set.