View-based 3-D Object Retrieval

Yue Gao, Qionghai Dai

• Publisher: Morgan Kaufmann
• Year: 2014
• Print ISBN: 9780128024195
• eBook ISBN: 9780128026236
• Topic: Computer Science
• Subtopic: Computer Graphics
• Index: Computer Science

Content-based 3-D object retrieval has attracted extensive attention recently and has applications in a variety of fields, such as, computer-aided design, tele-medicine, mobile multimedia, virtual reality, and entertainment. The development of efficient and effective content-based 3-D object retrieval techniques has enabled the use of fast 3-D reconstruction and model design. Recent technical progress, such as the development of camera technologies, has made it possible to capture the views of 3-D objects. As a result, view-based 3-D object retrieval has become an essential but challenging research topic.View-based 3-D Object Retrieval introduces and discusses the fundamental challenges in view-based 3-D object retrieval, proposes a collection of selected state-of-the-art methods for accomplishing this task developed by the authors, and summarizes recent achievements in view-based 3-D object retrieval. Part I presents an Introduction to View-based 3-D Object Retrieval, Part II discusses View Extraction, Selection, and Representation, Part III provides a deep dive into View-Based 3-D Object Comparison, and Part IV looks at future research and developments including Big Data application and geographical location-based applications.- Systematically introduces view-based 3-D object retrieval, including problem definitions and settings, methodologies, and benchmark testing beds- Discusses several key challenges in view-based 3-D object retrieval, and introduces the state-of-the-art solutions- Presents the progression from general image retrieval techniques to view-based 3-D object retrieval- Introduces future research efforts in the areas of Big Data, feature extraction, and geographical location-based applications

Part I

The Start

Introduction

This part, which serves as the introduction of the book, contains the first two chapters. In Chapter 1, the basic ideas and the background of 3-D object retrieval (3DOR) are introduced, including the comparison between model-based 3DOR (M3DOR) and view-based 3DOR (V3DOR), the main challenges of V3DOR, our contributions, and the structure of this book. In Chapter 2, we introduce the benchmark for 3DOR in recent years and the Shape Retrieval Contest (SHREC).

Chapter 1

Introduction

Abstract

3-D objects have become important multimedia data with various applications. The increasingly large scale of 3-D data necessitates the development of the 3DOR technique, which has garnered extensive research attention in recent years. This chapter briefly reviews this field, including the definition of 3DOR and the two general categories of 3DOR: M3DOR and V3DOR. We also summarize the main challenges of V3DOR. We have proposed a series of studies to target these challenges; these studies are summarized in this chapter, including view extraction, representative view selection, learning view weights, distance measurements, and object relevance learning. The structure of this book is also introduced in this chapter.

Keywords

3-D object retrieval

Challenges of 3DOR

View extraction

View selection

View weighting

Distance measure

Relevance learning

1.1 The Definition of 3DOR

Rapid advances in computer techniques and networks have led to large-scale multimedia data, which necessitate effective and efficient information retrieval. In the past several decades, multimedia information retrieval has been investigated extensively [1–3]. In recent years, computing graphics hardware and image processing techniques have made remarkable progress and 3-D technologies have been shown to be superior in various application fields, such as computer-aided design (CAD), medicine, virtual reality, and entertainment. For example, 3-D movies have become much more popular nowadays and their market is still increasing.

Traditional devices for collecting visual information, such as cameras and video recorders, are primarily used for 2-D information. Although 2-D information processing has been investigated for decades, 3-D data are superior for representing and processing stereo information.

The applications of 3-D technology can be traced back to the nineteenth century. In 1839, the English scientist Sir Charles Wheatstone invented the first stereoscope. Stereopsis is described by binocular vision, which was used to construct the stereoscope. Sir Charles Wheatstone explained that the impression of solidity was achieved by the combination of two individual views of the same objects, and these views were taken by our two eyes. Dr Patrick J. Hanratty, the father of computer-aided manufacturing (CAM), developed the first CAD software system in 1957, a numerical control programming tool that was also the first commercial CAM software. During that time, CAD modeling was still based on 2-D drafting. In 1965, a team led by Charles Lang from Cambridge University started conducting 3-D CAD modeling research. As a result, 3-D modeling became popular and has since been applied to various design tasks. In terms of entertainment, the first 3-D game “Night Driver” was released in 1976 by Atari. Nowadays, 3-D effects have become a central component of electronic games such as “World of War.”

With the wide applications of 3-D technology, 3-D data provide quick increases in both local data storage and online data storage from the Internet. The growing stores of data naturally increase the urgent need for effective and efficient 3DOR methods. Although 3-D modeling has been investigated for decades, it is still a high-cost and laborious task to model highly realistic 3-D models. According to a report from Gunn [4], only 20% of designs should be started from the very beginning in industrial design, while 40% of designs can be achieved by combining existing designs, and the remaining 40% of designs can be obtained by just revising existing designs. This percentage breakdown indicates that the reuse and revision of existing 3-D models can significantly improve model design performance, which necessitates the importance of 3-D object search in industry.

The main task of 3DOR can be defined as follows [5]:

Given a query object, define appropriate measures to automatically assess the similarity between any pair of 3-D objects based on a suitable notion of similarity.

Considering the requirements of 3DOR, the “content-based” method is the preferred technique: this method has been shown to be superior in multimedia information retrieval. In comparison with existing 1-D and 2-D multimedia information retrieval techniques, 3DOR is still in its infancy. Traditional text-based 3DOR methods which suffer from low efficiency and accuracy, cannot be effectively employed. One important issue is that 3-D objects contain both shape and appearance information, which is beyond the scope of text-based representation. Under these circumstances, extensive research efforts [6–8] have been dedicated to 3DOR, in which the content-based information has been exploited for intrinsic featu...

Table of contents

• Cover image
• Title page
• Table of Contents
• Copyright
• Acknowledgments
• Preface
• Part I: The Start
• Part II: View Extraction, Selection, and Representation
• Part III: View-Based 3-D Object Comparison
• Part IV: Conclusions and Future Work