Published Paper Details:

Facial expression recognition (FER) is an important component in improving human computer interaction through the ability of machines to recognize human emotions. The new improved FER systems have been using deep learning techniques, especially Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs). Real time emotion recognition is something that CNNs excel at as their spatial feature extraction powers them to be very good at extracting spatial features from static images. On the other hand, RNNs are aptly suited for temporal sequences, that is, for capturing of emotion progression with time that is crucial for video based FER. In this paper, we offer a comparison of CNNs and RNNs, with respect to their strengths, limitations, and where we think they are best applied. It also examines hybrid models that combine both structures, and thus presents a complete approach to exploiting the combined capabilities of both architectures. Our findings suggest that CNNs are well suited for fast, static image recognition, while RNNs are better suited for dynamic, sequence based tasks. The next step in future research should extend to making models more robust and integrating multimodal data to make more adaptive and accurate FER systems.