Recognition Performance of Facial Expression for the Face’s Partial Regions

With the rapid development of artificial intelligence, automatic facial expression recognition has been intensively investigated. However, it cannot maintain high accuracy of facial expression recognition due to face’s partial occlusion because most of facial expression recognition methods are designed based on the assumption that the entire face is visible. Therefore, the purpose of this study is to develop a method that does not degrade the accuracy of facial expression recognition even if a part of the face is occluded. In this paper, we investigate the accuracy of the facial expression recognition for only the region around the eyes using the CK+ dataset. The 3-D CNN and 2-D CNN with synthetic or subtracted eye images as the input image were adopted in the experiment The experimental results showed that the accuracy of facial expression recognition using the 3-D CNN or 2-D CNN with subtracted eye images were improved. Therefore, the temporal variations of facial expression are effective for the facial expression recognition using only the region around the eyes.