Modelling of Shapes without Landmarks
The complexity in variation that objects are provided with motivates to consider learning strategies when modeling their shape. This paper evaluates auto-associative neural networks and their application to shape analysis. Previously, such networks have been considered in connection with ‘point distribution models’ for describing two-dimensional contours in a statistical manner. This paper suggests an extension of this idea to achieve a more flexible model that is independent of landmarks.
KeywordsFeature Space Shape Space Trained Network Active Shape Model Shape Data
- 4.Michael A. Sipe, David P. Casasent, and Leonard M. Neiberg. Feature space trajectory representation for active vision. SPIE, 1997.Google Scholar