2RULE

FITTING  ACTIVE CONTOUR MODEL TO FACES IN 

VIDEOPHONE SCENES (Snake Algorithms, Neural Network, Digital Avatar,Deep Fake, Synthetic Media,Computer Vision,)

​

This thesis presents an alternative approach to  Teleconferencing Systems and refers to a system which uses a face sequence as input for animation of a wireframe model.  The system exploits a  general model of human face originally developed for realistic facial animation.  The visual observation is achieved by using active contour models as snakes which work reciprocally with a  physical model describing facial ex­pression. The active contours serve to track the nonrigid motions of facial features and solve the correspondence problem of 3-D  motion estimation. The 3-D rigid motion parameters are estimated by a  neural network that is trained using backpropagation algorithm. This system uses  wireframe models and  provides the imitation of a face  se­quence based upon the facial expression and  3-D  rigid  motion parameters estimation