An intelligent three-dimensional engineering shape searching system - effectiveness and efficiency

Kuiyang Lou, Purdue University

Abstract

Recently, traditional Computer-Aided Design (CAD) systems have been evolving into more integrated product developing systems. These advanced systems manage product information through their lifecycle, from conception, operation, to disposal. Searching 3D engineering shapes to reuse corporate intellectual capital has been triggering substantial changes for the product development in the data-intensive environment. The thesis is focused on search effectiveness and efficiency of searching 3D engineering shapes. A prototypical search system is designed and implemented as the test-bed for our concepts and algorithms. Search effectiveness is characterized by precision and recall; and search efficiency is evaluated by the ratio of the number of the visited data pages in a search operation to that of the index tree. A real database and synthetic databases created by random-number generator are employed for the research. Four feature vectors - principal moments, geometric parameters, moment invariants, and eigenvalues - are extracted to represent 3D shapes. The descending order of search effectiveness of using feature vectors are: principal moments, moment invariants, geometric parameters, and eigenvalues. The performance of an R-tree based index is investigated to improve the search efficiency. The results show that R-tree index significantly improves the efficiency of both real and synthetic databases. Multi-step search and relevance feedback are explored to improve search effectiveness. The effectiveness of using multi-step refinement is 51% higher than that of one-shot search using principal moment as feature vector. Two relevance feedback mechanisms - query reinterpretation (QR) and weight reconfiguration (WR) - are investigated. QR improves search performance by 56%, and WR 35%. To the best of our knowledge, the thesis is the first to build up the benchmark database and test the effectiveness of different feature vector systematically. The complementary application of our shape descriptors - feature vector and skeletal graph - is unique and promising. The 3D viewing interface is the first of such in content-based retrieval system for 3D shape search. In addition, the multi-step search strategy is first proposed and applied in the thesis to improve search effectives. This approach as well as relevance feedback significantly improves the effectiveness of 3D shape search.

Degree

Ph.D.

Advisors

Ramani, Purdue University.

Subject Area

Mechanical engineering|Computer science

Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server
.

Share

COinS