Facial expression classification by temporal template features

In this paper, we proposed the facial expression recognition framework for estimating the emotional states by measuring the variations of facial activities under the context of human-machine interaction. The state of the art researches frequently estimate the facial expression by using a still image, which often mistaken recognize the input image as another emotion expression. In this paper, we deal with the problem by modeling the temporal template from the facial image sequences by using Motion History Image (MHI) or Cumulative Change of Feature (CCF), then we apply these features into the conventional linear classifier (EMC) or non-linear classifiers (KEMC and KNN). As a result, the temporal template can describe the facial expression with the fix number of feature dimensions regardless to the expression duration. Implicitly, MHI describes the order of the actions, while CCF displays the facial muscle activation levels.