Emotion Recognition in Human Thermal Images with Artificial Intelligence Technology

This paper presents a case study related to emotion recognition based on human thermal image processing. Three states are considered for human faces: normal, sad, and happy. The thermal images are pre-processed for important feature selection and extraction by using the random forest (RF) classifier. All and selected features set is used as input to a feedforward backpropagation neural network (FFBPN), with binary outputs. The input images are organized in pairs, i.e., normal with sad and normal with happy. The preliminary results show particularly good classification accuracy.