Cet ouvrage fait partie de la bibliothèque YouScribe
Obtenez un accès à la bibliothèque pour le lire en ligne
En savoir plus

Segmentation of color images for interactive 3D object retrieval [Elektronische Ressource] / vorgelegt von José Pablo Alvarado Moya

De
212 pages
Segmentation of color imagesfor interactive 3D object retrievalVon der Fakultat fur Elektrotechnik und Informationstechnikder Rheinisch-Westfalischen Technischen Hochschule Aachenzur Erlangung des akademischen Grades einesDoktors der Ingenieurwissenschaften genehmigte Dissertationvorgelegt vonDiplom-IngenieurJose Pablo Alvarado Moyaaus San Jose, Costa RicaBerichter: Universit atsprofessor Dr.-Ing. Karl-Friedrich KraissUniversit atsprofessor Dietrich Meyer-EbrechtTag der mundlic hen Prufung: 02. Juli 2004Diese Dissertation ist auf den Internetseiten der Hochschulbibliothek online verfugbara mis queridos padres(to my parents)AcknowledgmentsI extend the most sincere appreciation to all people and organizations who made thisthesis possible. My deepest gratitude goes to my advisor Prof. Dr.-Ing. K.-F. Kraiss,whoo eredmetheinvaluableopportunityofworkingasaresearchassistantathisChairandgavemethefreedomtoexplorethefascinatingworldofimageanalysis. Hissupportand guidance have been fundamentally important for the successful completion of thisdissertation. IamalsoindebtedtomysecondexaminerProf.Dr.-Ing.D.Meyer-Ebrechtfor his interest in my work.2TotheHeinz-NixdorfFoundationforsupportingtheprojectsAxon andAxiom,inwhichscope the current research work took place, and to Stei GmbH for providing such niceobject sets.To the DAAD for providing the nancial support for the rst stage of my studies inGermany: they made it possible to obtain my Dipl.-Ing.
Voir plus Voir moins

Segmentation of color images
for interactive 3D object retrieval
Von der Fakultat fur Elektrotechnik und Informationstechnik
der Rheinisch-Westfalischen Technischen Hochschule Aachen
zur Erlangung des akademischen Grades eines
Doktors der Ingenieurwissenschaften genehmigte Dissertation
vorgelegt von
Diplom-Ingenieur
Jose Pablo Alvarado Moya
aus San Jose, Costa Rica
Berichter: Universit atsprofessor Dr.-Ing. Karl-Friedrich Kraiss
Universit atsprofessor Dietrich Meyer-Ebrecht
Tag der mundlic hen Prufung: 02. Juli 2004
Diese Dissertation ist auf den Internetseiten der Hochschulbibliothek online verfugbara mis queridos padres
(to my parents)Acknowledgments
I extend the most sincere appreciation to all people and organizations who made this
thesis possible. My deepest gratitude goes to my advisor Prof. Dr.-Ing. K.-F. Kraiss,
whoo eredmetheinvaluableopportunityofworkingasaresearchassistantathisChair
andgavemethefreedomtoexplorethefascinatingworldofimageanalysis. Hissupport
and guidance have been fundamentally important for the successful completion of this
dissertation. IamalsoindebtedtomysecondexaminerProf.Dr.-Ing.D.Meyer-Ebrecht
for his interest in my work.
2TotheHeinz-NixdorfFoundationforsupportingtheprojectsAxon andAxiom,inwhich
scope the current research work took place, and to Stei GmbH for providing such nice
object sets.
To the DAAD for providing the nancial support for the rst stage of my studies in
Germany: they made it possible to obtain my Dipl.-Ing., which undeniably lead me into
the next stage as a research assistant. To the ITCR for ensuring me a job after all these
years in Germany.
I want to thank my colleagues Dr. Peter Walter and Dr. Ingo Elsen who helped me
with my rst steps in the world of computer vision, and especially to my project col-
leagues Jochen Wickel, Dr. Thomas Krug er and Peter D or er for providing such a rich
and fruitful environment for developing new ideas. To all my colleagues at the Chair
of Technical Computer Science for the healthy, symbiotic collaboration creating and
maintaining the LTI-Lib, which is becoming a great contribution to the research world
in image understanding.
To my student researchers Stefan Syberichs, Thomas Rusert, Bastian Ibach, Markus
Radermacher, Jens Rietzschel, Guy Wafo Moudhe, Xinghan Yu, Helmuth Euler for
their collaborative work in the image processing part of our projects, and specially to
Axel Berner, Christian Harte, and Frederik Lange for their excellent work and valuable
support.
Many a thanks are due to all my proofreaders Suat Akyol, Rodrigo Batista, Manuel
Castillo,MichaelH ahnel,LarsLibuda,whoreviewedsomeofthechapters,butespecially
to Peter Dor er, who patiently reviewed the whole thesis. I also want to thank Lars,
MichaelandPeterforthe“proof-listening”inthepreparationofthepublicpresentation.I am especially indebted to my dear friends Suat, Rodrigo, Jana, Manuel and Martha,
and to my family in Costa Rica for their huge support during the hard times while
preparing this thesis.
This work is dedicated to my mother and to the memory of my father. They always
encouragedmetogivemybestandinfusedmewithboundlesscuriosityandpersistence.
Thank you for your absolute con dence in me and for your understanding and support
for my pursuits.
Jose Pablo Alvarado Moya
Aachen, 02.07.2004Contents
List of Figures v
List of Tables ix
Glossary xi
List of symbols and abbreviations xiii
1 Introduction 1
1.1 Motivation and Problem Statement . . . . . . . . . . . . . . . . . . . . . 4
1.2 Analysis Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2.1 Marr’s Metatheory . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2.2 Computational Level of Vision Systems . . . . . . . . . . . . . . . 7
1.3 Goal and Structure of this Work . . . . . . . . . . . . . . . . . . . . . . . 8
2 Visual Object Retrieval 11
2.1 Task of Object Retrieval Systems . . . . . . . . . . . . . . . . . . . . . . 12
2.2 Taxonomy of Object Retrieval Systems . . . . . . . . . . . . . . . . . . . 13
2.2.1 Architecture of Object Retrieval Systems . . . . . . . . . . . . . . 14
2.2.2 Object Set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.2.3 Nature of Images . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.2.4 Scene Composition . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.2.5 Knowledge Type . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.3 Assessment of Object Retrieval Systems . . . . . . . . . . . . . . . . . . 24
2.4 Experimental Set-up . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
iContents
3 Foundations of Image Segmentation 29
3.1 De nition of Segmentation . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.2 Image-based Segmentation . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.2.1 Feature-space Approaches . . . . . . . . . . . . . . . . . . . . . . 36
3.2.2 Image-domain-based Approaches . . . . . . . . . . . . . . . . . . 38
3.2.3 Hybrid Methods. . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.2.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.3 Surface-based Segmentation . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.4 Object-based Segmentation . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.5 Segmentation in Current Object Recognition Systems . . . . . . . . . . . 49
3.6 Framework for Segmentation in Object Retrieval Applications . . . . . . 52
3.7 Evaluation of Segmentation Algorithms . . . . . . . . . . . . . . . . . . . 53
3.7.1 Evaluation Using the Pareto Front . . . . . . . . . . . . . . . . . 55
3.7.2 Fitness Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4 Image-based Segmentation 61
4.1 Mean-Shift Segmentation . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.2 Watershed-based Segmentation . . . . . . . . . . . . . . . . . . . . . . . 65
4.2.1 Color Edgeness . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4.2.2 Reducing Over-Segmentation . . . . . . . . . . . . . . . . . . . . 68
4.3 k-Means-based Segmentation. . . . . . . . . . . . . . . . . . . . . . . . . 74
4.4 Adaptive Clustering Algorithm . . . . . . . . . . . . . . . . . . . . . . . 76
4.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
5 Surface-Based Segmentation 87
5.1 Physical Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
5.2 Perceptual Organization . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
5.3 Application Dependent Cues . . . . . . . . . . . . . . . . . . . . . . . . . 90
5.3.1 Encoding Color Information . . . . . . . . . . . . . . . . . . . . . 91
5.3.2 Encoding Positional Information . . . . . . . . . . . . . . . . . . . 92
5.3.3 Combining Cues with Bayesian Belief Networks . . . . . . . . . . 93
iiContents
5.3.4 Example: Detection of an Homogeneous Background . . . . . . . 95
5.4 Example: Separation of Hand Surfaces and Skin-Colored Objects . . . . 100
5.4.1 Selection of the Skin-Color Model . . . . . . . . . . . . . . . . . . 101
5.4.2 Color Zooming . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
5.4.3 Probabilities for Color Labels . . . . . . . . . . . . . . . . . . . . 103
5.4.4 Positional Information . . . . . . . . . . . . . . . . . . . . . . . . 104
5.4.5 Combining All Information Cues . . . . . . . . . . . . . . . . . . 105
5.4.6 Seed Selection and Growing Stages . . . . . . . . . . . . . . . . . 106
6 Object-Based Segmentation 109
6.1 Recognition Based on Global Descriptors . . . . . . . . . . . . . . . . . . 110
6.1.1 Global Descriptors . . . . . . . . . . . . . . . . . . . . . . . . . . 110
6.1.2 Classi cation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
6.1.3 Recognition Experiments with Global Descriptors . . . . . . . . . 117
6.2 Recognition Based on Local Descriptors . . . . . . . . . . . . . . . . . . 119
6.2.1 Location Detection . . . . . . . . . . . . . . . . . . . . . . . . . . 120
6.2.2 Descriptor Extraction. . . . . . . . . . . . . . . . . . . . . . . . . 125
6.2.3 Classi cation of Local Descriptors . . . . . . . . . . . . . . . . . . 127
6.2.4 Assignment of Object Labels to Regions . . . . . . . . . . . . . . 131
6.3 Object-based Segmentation Module . . . . . . . . . . . . . . . . . . . . . 132
6.4 Recognition Experiments with Multiple Objects . . . . . . . . . . . . . . 135
7 Conclusion and Outlook 139
7.1 Summary of the Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
7.2 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
Bibliography 145
A Pareto Envelope-based Selection Algorithm 161
B Optimal region pair selection 165
C Fast Relabeling 169
iiiContents
D Images for Evaluation of Image-Based Segmentation 173
E Test-set P17 175
F Test-sets S25 and S200 177
G Computation of the Hessian Matrix 179
Index 185
iv

Un pour Un
Permettre à tous d'accéder à la lecture
Pour chaque accès à la bibliothèque, YouScribe donne un accès à une personne dans le besoin