Unusual Expression Detection in Multiple Scenarios Based on One-Class SVM
Xiaopu Wang, Jinwei Deng, Chao Qi · 2021 IEEE Intl Conf on Dependable, Autonomic and Secure Computing, Intl Conf on Pervasive Intelligence and Computing, Intl Conf on Cloud and Big Data Computing, Intl Conf on Cyber Science and Technology Congress (DASC/PiCom/CBDCom/CyberSciTech) · 2021
Human facial expressions are often abnormal when they have a sudden illness. Time will be saved for taking emergency measures if we can catch the anomalies in time. But there is not enough abnormal data for the model to learn. It is difficult to evaluate unknown anomalies even if a certain amount of abnormal data is obtained. One-Class Support Vector Machine(One-Class SVM) can solve this problem by calculating the boundary of normal/abnormal data without relying on any independent abnormal data. This paper analyzed the anomalies of facial expression changes which are presented in multi-scene videos recorded by camera. The whole process is divided into three stages: face region segmentation, feature extraction and anomaly classification. We used multi-task convolutional neural network (MTCNN) to detect and align faces, and used FaceNet to extract facial features. In the process of anomaly classification, the performance of One-Class SVM under different kernels of Linear, Polynomial, RBF, and Sigmoid was recorded.