Fusing dynamic deep learned features and handcrafted features for facial expression recognition

Abstract

The automated recognition of facial expressions has been actively researched due to its wide-ranging applications. The recent advances in deep learning have improved the performance facial expression recognition (FER) methods. In this paper, we propose a framework that combines discriminative features learned using convolutional neural networks and handcrafted features that include shape- and appearance-based features to further improve the robustness and accuracy of FER. In addition, texture information is extracted from facial patches to enhance the discriminative power of the extracted textures. By encoding shape, appearance, and deep dynamic information, the proposed framework provides high performance and outperforms state-of-the-art FER methods on the CK+ dataset.