We propose a segmentation technique adapted to objects composed of several regions with gray-level fluctuations described by different probability laws. This approach is based on information theory techniques and leads to a multiregion polygonal snake driven by the minimization of a criterion without any parameters to be tuned by the user. We demonstrate the improvements obtained with this approach as well as its low computational cost. This approach is compatible with applications such as object recognition and object tracking with nonrigid deformation in images perturbed by different types of optical noise.
© 2004 Optical Society of America
Frédéric Galland and Philippe Réfrégier, "Information-theory-based snake adapted to multiregion objects with different noise models," Opt. Lett. 29, 1611-1613 (2004)