<![CDATA[Facial emotion recognition (FER) is a research topic that focuses on the analysis of human facial expressions. There are many FER research has been conducted on single images or photo. Emotion analysis on single images has many disadvantages compared to dynamic image sequences or videos. This is due to human emotions or expressions within a certain time. The classification of emotions becomes complicated when considering different emotions. There are some people who are very expressive, there are some people who have low or moderate expressions. Predictions of emotion with variety intensities has decresed error due to data sets that provide only a few emotions intensities. Data annotation is a major problem in recognition fields that require a lot of time and effort to annotate new data. This study aims to find information about facial emotions with emotional intensity from subtle to sharp in a sequence images or videos. The dataset will be trained using Convolutional neural network by augmentation to add data annotations. The proposed method was evaluated using the public BP4D-Spontaneous dataset. The evaluation results show that the average emotion recognition in video sequences using the holdout method is 18%. Evaluation of the loss function parameter shows overfitting where the curve generalization gap is too high. The last evaluation is the evaluation of the emotion class between the real class and the prediction class in 14.28%. This shows that the classification of emotion recognition in dynamic image sequences is quite low.]]>
Copyrights © 2020
