Global fitting of a facial model to facial features for model-based video coding

Finding faces, locating important facial features and model fitting are required in a range of applications including model based video coding. In this paper we present a technique for locating important facial features in head and shoulders colour images. A fast and effective method of detecting skin, while excluding eyes, using PCA transformed CIE-Lab colour space, is described. Candidate eye regions are then processed using region growing and eigen images using the distance from feature space approach. False matches are eliminated using facial geometry constraints. Nose and mouth features are also located using eigen images in search regions derived from the eye positions. Further facial characteristics are identified and automated fitting of a global head model is achieved using a genetic algorithm based approach. Results from applying our technique to a large database of facial images (XM2VTS) are presented.