Facial Emotion Recognition Using Transfer Learning
Shamoil Shaees, Hamad Naeem, Muhammad Arslan, Muhammad Rashid Naeem, Syed Hamza Ali, Hamza Aldabbas
Abstract
Facial Emotion Recognition is considered very significant for Human Computer Interaction (HCI) and they play a vital role in everyday human life. In recent years Convolutional Neural Network (CNNs) has become very popular among researcher for image analysis, because CNNs have generated remarkable results. However, CNNs needs a lot of data to train from scratch. This problem has been addressed by several researchers who have trained CNNs with millions of images, this training knowledge can also be used in a different task which is known as Transfer Learning. AlexNet is one of the best pre-trained CNNs. Our work reflects a brief comparison between modern pre-trained CNNs and using transfer learning with classification approach like Support Vector Machine (SVM), generally known as hybrid classifier. The testing has been done on two very popular expression database Cohn-Kanade+ (CK+) database and Natural Visible and Infrared Expression (NVIE) database. Results clearly depicts that pre-trained CNNs produces better result than handcrafted approaches.