Graph-Cut Based Face Reconstruction and Emotions Reproduction

Charlotte Ghys


Abstract

This talk will present my work in France on stereo face reconstruction and on emotions reproduction.
The approach for face reconstruction is based on graph cut. Graph cut is a fast global and optimal minimization method that solves Markov Random Field problems. It provides so a smooth surface reconstruction without outliers and can be use for a lot of different field in computer vision (segmentation, mosaicing, multiview reconstruction ...).

The emotion reproduction is another task of the Face Research Area. Our face model is an adapted version of the MPEG-4 Model.The goal of this method is to reproduce in 3D the 6 basic emotions using a linear or non-linear auto regressive model.
I will conclude the talk explaining my work in IRIS Lab.


Maintained by Adit Sahasrabudhe