Abstract:
Online instruction has been increasingly popular because the accessibility to content via this method is easier than conventional instruction. However, there are some limitations about user verification and learning evaluation in terms of emotion expression because a learner does not interact directly with an instructor. Therefore, face recognition and face expression recognition are suitable techniques to solve this problem through the analysis of learners face images while an online course.
For face recognition, the researcher employed a convolutional neural network and NN4 model from FaceNet research on data set of 65 students at the Department of Computer Education, each with 3 images, using the maximum threshold. The results showed that the validation rate was at 100% and the false accept rate was at 21.92%. If the number of images is less than 3, the results showed that threshold at 0.57 was a level that yielded the best performance of face recognition of CED2017 data set. The validation rate was at 75.89% and the false accept rate was at 17.84%.
For face expression recognition, the researcher employed a convolutional neural network with FER2013 data set divided into three categories, namely negative, normal, and positive emotions. The data set was first passed through 18 image processing methods before being trained with 8 different models for comparisons. The comparative results showed that 3CNNs New Group Alignment with Flip had the accuracy of 84.22% and SEARCH data set had the accuracy of 92.5%.