Mobilní komunikační aplikace s automatickou klasifikací emocí z obličeje

Abstract

This thesis explores the possibilities of human facial emotion classification using camera images on Android devices. Neural network architectures optimized for the lower power resources of mobile devices were compared. Training and evaluation of each neural network model was performed within the Keras API of the TensorFlow library, followed by conversion to the TensorFlow Lite standard to reduce memory and computational requirements. The entire process, from face detection to emotion classification, works in real time. The functional and optimized solution serves as an extension of an open source communication application implementing the XMPP protocol. Both users within the same chat have an overview of each other's real emotions.