Abstract:
Nowadays, there are many 3D objects in digital libraries. Many approaches have been proposed in order to improve the speed and effectiveness of the retrieval. However, these approaches are not sufficient compared to the text-based retrieval. Because most of the existing 3D retrieval solutions are not support the similarity between part of object, that is partial similarity. Therefore, it is unable to query by the part of object and also cannot diagnose the element of object as text based retrieval. In this thesis, we present an algorithm for partial shape retrieval on a collection of 3D polygonal meshes. The proposed algorithm is invariant against rigid transformations and robust against different pose by using structure properties and geometric properties to represent shapes. Structure property is represented by a Reeb graph which uses an integral geodesic distance as a Morse function, whereas geometric property is represented by a Pose invariant Shape Signature. The main idea is to use Reeb graph for decomposing shape into many meaningful sub parts. Then describe each sub part with geometric property. The similarity is computed based on the Approximate Maximum Common Sub graph for matching each subpart between query shape and other while preserving topology. We evaluate our algorithm on various different model classes and deformation. The experimental results indicate better accuracy compared to the previous method in the case of deformable object. We have conclude that our approach is fast and sufficient for practical use. However, our algorithm is not suitable for concave objects and convex hull objects. The computational cost of the algorithm is O(n log n) for describing a shape and O(m⁴) for each matching with n is the number of vertices in mesh and m is the number of node in Reeb graph. The Mean Average Precision of this algorithm is about 0.348.