Beyond Traditional Emotion Recognition

Abstract

This thesis investigates into the problem of recognizing human emotions in a near real life environment. This goes beyond the traditional works on emotion recognition which perform Emotion Recognition (ER) in a controlled environment on a frontal face, with sufficient illumination, optimal face size, minimal head motion and so on. Specifically the following ways are proposed to deal with difficulties in the near real world data: 1. Using 3D instead of conventional 2D information for recognizing facial expressions which is one of the modalities contributing to ER, and 2. Fusing information from multiple modalities. An application of Emotion Recognition is presented in the form of automated personality assessment using multimodal features including those extracted based on emotion recognition.