Automatic Detection of Facial Landmarks from AU-coded Expressive Facial Images

The present aim was to develop a fully automatic feature-based method for expression-invariant detection of facial landmarks from still facial images. It is a continuation of our earlier work where we found that some certain muscle contractions made a deteriorating effect on the feature-based landmark detection especially in the lower face. Taking into account this crucial facial behavior, we introduced improvements to the method that allowed facial landmarks to be fully automatically detected from expressive images of high complexity. In the method, information on local oriented edges was utilized to compose edge maps of the image at two levels of resolution. The landmark candidates resulted from this step were further verified by edge orientation matching. We used knowledge on face geometry to find the proper spatial arrangement of the candidates. The results obtained demonstrated a high overall performance of the method while testing a wide range official displays.