Sale!
Placeholder

Object Recognition Using An Interest Operator

10,000 3,000

Topic Description

Summary
This project is based in the field of computer vision and aims to re-implement and evaluate an
object recognition system designed by Bastian Leibe and Bernt Schiele in the publication
“Interleaved Object Categorization and Segmentation”. Drawing on research from other
techniques, the goal is to improve on methods within this system during implementation and
further extend the system by tackling the problem of recognising multiple object categories.
The project heads towards producing a solution to the problem of unsupervised learning and a
system that has true artificial intelligence.

Contents
1 Project Introduction 1
1.1 Aim 1
1.2 Objectives 1
1.3 Minimum Requirements (deliverables) 2
1.4 Methodology 2
1.5 Relevance to My Degree Programme 3
2 Background Research 5
2.1 A Survey of Recognition 5
2.2 Leibe and Schieles’ System, a “local approach” 9
2.3 Harris, Finding interesting Features 11
2.4 Appropriate tools 12
2.5 Summary 13
3 Implementation 14
3.1 The Learning Stage 14
3.11 Implementing Harris 15
3.12 Clustering interest patches 19
3.13 The Codebook 22
3.2 The Recognition Stage 24
3.21 Finding the objects centre 25
3.22 Back-projection 27
3.23 Refining the hypothesis 27
3.24 The complete recognition system 29
4 Evaluating the System 31
4.1 Test Data 31
4.2 Defining a test strategy 33
4.3 Results 35
4.31 The ROC curve 36
4.32 The Recall-Precision curve 37
– iv –
4.4 Conclusion 38
5 Extending the Solution 39
5.1 A simple method 40
5.2 Thinking in dimensions 40
5.21 Multidimensional scaling 41
5.22 Decomposing the affinity matrix 43
5.23 Principal Components Analysis 45
5.3 Integrating into the system 47
6 Evaluating the project 49
6.1 Achievements 49
6.2 Future work 50
Bibliography 51
Appendix A Project Reflections 53
Appendix B Project Management

GET COMPLETE PROJECT