Facial expression recognition using distance importance scores between facial landmarks

Abstract

In this paper, we present a feature extraction approach for facial expressions recognition based on distance importance scores between the coordinates of facial landmarks. Two audio-visual speech databases (CREMA-D and RAVDESS) were used in the research. We conducted experiments using the Long Short-Term Memory Recurrent Neural Network model in a single corpus and cross-corpus setup with different length sequences. Experiments were carried out using different sets and types of visual features. An accuracy of facial expression recognition was 79.1% and 98.9% for the CREMA-D and RAVDESS databases, respectively. The extracted features provide a better recognition result compared to other methods based on the analysis of facial graphical regions.