An Experiment on 3D Face Model Adaptation using the Active Appearance Algorithm
Jörgen Ahlberg · 2001
This report describes our first implementation and experiment on adapting a 3D wirefrane model to a face in an input image using the Active Appearance Algorithm. The results are promising, and it is our conclusion that we should continue on this track in our task to create a real-time model-based coder. 1. Introduction Our goal is to find and track a human face and its features in a video sequence. We want to do this by adapting, in each frame, a wireframe model to the face in the image. This should be done accurately enough to allow a realistic-looking facial animation to be created from the extracted face model parameters, and it should also have real-time performance. We use the CANDIDE-3 [4] model since it is simple, but yet able to represent all the parameters we wish to extract. This report describes our first try to use the Active Appearance Algorithm to adapt the model to the image. The concept of Active Appearance Models (AAMs) was introduced a few years ago [2], and ha...