Emotion recognition from facial expressions based on multi-level classification

Abstract

Emotion plays an imperative role in non-verbal communication. The emotion recognition applications have demonstrated their capabilities in the next generation human-machine interactive system. In this paper, we present a multilevel classification framework for human emotion recognition from facial images. The proposed framework for emotion recognition consists of two phases: face processing and emotion classification. In the first phase, we have described the whole methodology used for face localisation and facial feature detection. The second phase, a multi-level classification approach is used for emotion recognition. In the proposed approach, the classification accuracy of principal component analysis (PCA) at level 1 is boosted by support vector machines (SVMs) at level 2, which is appropriate for representing the expression patterns having large intraclass variations. Experimental results demonstrate that the proposed approach can successfully recognise facial emotion with 94% recognition rate.