Access Type

Open Access Dissertation

Date of Award

January 2011

Degree Type


Degree Name



Computer Science

First Advisor

Jing Hua


This study aims to classify different deformations based on the shape space concept. A shape space is a quotient space in which each point corresponds to a class of shapes. The shapes of each class are transformed to each other by a transformation group preserving a geometrical property in which we are interested. Therefore, each deformation is a curve on the high dimensional shape space manifold, and one can classify the deformations by comparison of their corresponding deformation curves in shape space. Towards this end, two classification methods are proposed.

In the first method, a quasi conformal shape space is constructed based on a novel quasi-conformal metric, which preserves the curvature changes at each vertex during the deformation. Besides, a classification framework is introduced for deformation classification. The results on synthetic and real datasets show the effectiveness of the metric to estimate the intrinsic geometry of the shape space manifold, and its ability to classify and interpolate different deformations.

In the second method, we introduce the medial surface shape space which classifies the deformations based on the medial surface and thickness of the shape. This shape space is based on the log map and uses two novel measures, average of the normal vectors and mean of the positions, to determine the distance between each pair of shapes on shape space.

We applied these methods to classify the left ventricle deformations. The experimental results shows that the first method can remarkably classify the normal and abnormal subjects but this method cannot spot the location of the abnormality. In contrast, the second method can discriminate healthy subjects from patients with cardiomyopathy, and also can spot the abnormality on the left ventricle, which makes it a valuable assistant tool for diagnostic purposes.