Isometry-invariant Surface Matching: Numerical Algorithms and Applications

Similarity is one of the most important abstract concepts in the human perception of the world. For example, we encounter it every day during our interaction with other people whose faces we recognize. In computer vision and pattern recognition, shape similarity is one of the most fundamental and largely open problem. With slight exaggeration, we can say that all pattern recognition problems boil down to giving a quantitative interpretation of similarity between objects. The definition of similarity is, to a large extent, a semantic question. For example, judging the similarity of faces, one may say that two human faces are similar if they have a common skin tone, while someone else would require the identity of the geometric structure of facial features like the eyes, the nose, and the mouth. Since there is no unique definition of similarity, every class of objects and every problem require a specific, problem-dependent similarity criterion. This thesis is dedicated to the question of similarity of non-rigid shapes. Such shapes appear at all scales in Nature – from our body organs to its microscopic components like protein molecules. In pattern recognition and computer vision, non-rigid shapes are often encountered in many important applications. The richness of the possible deformations and a large number of degrees of freedom of non-rigid shapes makes their analysis a great challenge. Deformations of many natural objects can be modeled as isometries, preserving the intrinsic geometric properties of the shapes. For example, deformations of human and animal bodies, our face, body tissues and organs can be approximated by isometries. 1 T ec hn io n C om pu te r Sc ie nc e D ep ar tm en t P h. D . T he si s P H D -2 00 704 2 00 7