Abstract
This paper aims for swapping face images for other persons' face images based on time-sequential coherence facial expression transformations to protect portrait rights and privacy. Facial feature points are extracted by ASM (Active Shape Model) to obtain face parts those are a face base, eyes and a mouth. The extracted parts are distinguished by AU (Action Units) corresponding to "Happiness" and "Sadness." A slope and a distance are calculated using facial feature points for searching the parts corresponding to the facial expression. The parts positions are estimated by summations of errors at all feature points belonging to the each part. They are allocated when the errors are minimized. As a result, face swapping images based on time-sequential coherence facial expression transformations are generated.
