A comparative study of active appearance model annotation schemes for the face
Amrutha Sethuram, Karl Ricanek, Eric K. Patterson · 2010
Active appearance models (AAM), a group of flexible deformable models, have been widely used in various applications such as, object tracking, medical image segmentation and synthesis. AAMs are statistical models which model the shape and texture of an object. There has been much published work in this field to improve the speed and fitting accuracy. However, there has not been any significant study related to the quantity and selection of annotation points (landmarks) used to define the object and its texture. This paper proposes four different annotation schemes used for modeling the human face and evaluates each scheme in regard to reconstructing face images. In pursuit of choosing a particular annotation scheme for age progression and synthesis, this paper presents qualitative and quantitative methods for evaluation.