Use this URL to cite or link to this record in EThOS:
Title: Curve-based shape matching methods and applications
Author: Bouganis, Alexandros
ISNI:       0000 0004 2687 1752
Awarding Body: Imperial College London
Current Institution: Imperial College London
Date of Award: 2009
Availability of Full Text:
Access from EThOS:
Access from Institution:
One of the main cues we use in our everyday life when interacting with the environment is shape. For example, we use shape information to recognise a chair, grasp a cup, perceive traffic signs and solve jigsaw puzzles. We also use shape when dealing with more sophisticated tasks, such as the medical diagnosis of radiographs or the restoration of archaeological artifacts. While the perception of shape and its use is a natural ability of human beings, endowing machines with such skills is not straightforward. However, the exploitation of shape cues is important for the development of competent computer methods that will automatically perform tasks such as those just mentioned. With this aim, the present work proposes computer methods which use shape to tackle two important tasks, namely packing and object recognition. The packing problem arises in a variety of applications in industry, where the placement of a set of two-dimensional shapes on a surface such that no shapes overlap and the uncovered surface area is minimised is important. Given that this problem is NP-complete, we propose a heuristic method which searches for a solution of good quality, though not necessarily the optimal one, within a reasonable computation time. The proposed method adopts a pictorial representation and employs a greedy algorithm which uses a shape matching module in order to dynamically select the order and the pose of the parts to be placed based on the “gaps” appearing in the layout during the execution. This thesis further investigates shape matching in the context of object recognition and first considers the case where the target object and the input scene are represented by their silhouettes. Two distinct methods are proposed; the first method follows a local string matching approach, while the second one adopts a global optimisation approach using dynamic programming. Their use of silhouettes, however, rules out the consideration of any internal contours that might appear in the input scene, and in order to address this limitation, we later propose a graph-based scheme that performs shape matching incorporating information from both internal and external contours. Finally, we lift the assumption made that input data are available in the form of closed curves, and present a method which can robustly perform object recognition using curve fragments (edges) as input evidence. Experiments conducted with synthetic and real images, involving rigid and deformable objects, show the robustness of the proposed methods with respect to geometrical transformations, heavy clutter and substantial occlusion.
Supervisor: Shanahan, Murray Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral