Abstract:

Facial expression recognition has a crucial role in communication. Computerized facial expression recognition systems have been developed for many purposes. People's faces can have occlusions because of scarves, facial masks, etc. in cases such as cold weather conditions or Covid-19 pandemic conditions. In this case, facial expression recognition can be challenging for automated systems. This study classifies facial images containing only the eyebrow and eye regions over six expressions with a deep learning-based approach. For this purpose, Radboud Face Database images have been used after cropping the area that includes eye and eyebrow regions. Some popular pre-trained networks have been trained and tested using the transfer learning approach. The Vgg19 pre-trained network achieved 91.33% accuracy over the six universal facial expressions. The experiments show that automated facial expression recognition can be applied with high performance by looking at the region containing eyes and eyebrows

Keywords: