Piecewise DM : a Locally Controllable Deformable Model S

Deformable model methods (DM) constitute a class of segmentation techniques used to delineate the boundary of objects in the image. They represent a promising platform for the implementation of interactive segmentation because they allow for the elegant combination of information derived from the image data, constraints expressing prior knowledge about the boundary of interest and information provided by the user. When adopting existing DM to actually implement an interactive method, several limiting factors were encountered, motivating the development of a new DM. In this text we start identifying the basic elements of a DM, showing examples found in the literature. Next we de ne requirements posed on a DM to address the needs of interactive segmentation, reviewing eligible methods. An extension to the class of DM addressing these requirements is proposed, namely Piecewise DM, providing a general framework for the implementation of a exible and controllable DM with a larger rate of success in real segmentation problems. To conclude, we illustrate how the new method is applied to a complex segmentation task.