A0323
Title: Statistics on locally parametrized shapes via discrete swept skeletal representations
Authors: Joern Schulz - University of Stavanger (Norway) [presenting]
Mohsen Taheri Shalmani - University i Stavanger (Norway)
Abstract: Effective statistical shape analysis, such as detecting local dissimilarities of image objects, depends on the underlying shape representation and how the representation establishes local correspondence between the objects in a population. In opposite to most shape representations that are based either on non-invariant spatial geometrical object properties (GOPs) or on extrinsic GOPs, we propose a novel skeletal shape representation by defining local coordinate systems (fitted frames) at each location on the object skeletal and thereby defining intrinsic GOPs that are invariant to rigid transformations (no need for object pre-alignment) which supports statistical analysis. We will discuss how we can fit a new type of skeletal representation, namely the discrete swept skeletal representation, to a globular object. Finally, we will study the introduced shape representation based on simulated data and data from the ParkWest study. We compare the shapes of the left hippocampi of patients with Parkinson's disease versus a healthy control group.
