Online Kernel-Based Structured Output SVM for Early Expression Detection

摘要: As a key component of human–computer intelligent interaction and many real-world applications, the real-time property facial expression recognition is especially important. However, result conventional video-based approaches can not be given until entire video finished. In this letter, we deal with early detection, which aims to identify as possible before its ending. This relatively new challenging problem. Max-margin event detector (MMED) well-known framework, make detection. linearity restricts applications. We thus introduce kernel learning model nonlinear structure complex data distribution. Moreover, further reformulated in an online setting address streaming videos. The high retraining cost large memory requirement MMED are significantly reduced. addition, employ AlexNet architecture comparison mid-level features. Experiments on two popular datasets demonstrate both effectiveness efficiency proposed method.