Shape Matching in Visual Data

Acknowledgements First and foremost, I would like to thank my advisors Professor Andrea Massa and Professor Francesco G.B. De Natale who taught me a lot about research and life in general. They were always there to provide everything not only to support my research but also to help me adjust myself in new country as Italy. Without their guidance and support, this thesis would not have been completed. I also wish to thank the other members of my committee – Professor Fabrizio Granelli, Professor Piero Castoldi, and Professor Roberto Rinaldo. Thank you for agreeing to serve on my committee. I owe thanks to many people in ICT who help me out of troubles whenever I have. I thank all of my colleagues, my friends for being mutual well-wishers from whom I can gain a lot of things related to research and life. Especially, I lovely thank to my angle " Tròn " for giving me all her love without any hesitation, though I am usually a bad guy. I lovely prefer to call her " Tròn " (i.e. the circle) because: when you love someone, draw a circle around their name instead of a heart because hearts can be broken but circles never end. Finally, I give my deepest gratitude to my family for their endless support and encouragement. Abstract Nowadays, Content-based Image-Video Retrieval (CBIVR) plays an important role in an extraordinary number of multimedia applications which serve human society. In this manner, a shape-based image matching is a fundamental ingredient for CBIVR. Since a generic shape-based retrieval scheme working in every situation can be quite difficult to implement due to the presence of complex or cluttered environments, occlusions problems, deformations , etc., this thesis investigates to deal with such problems and focuses on two parts: The first part researches how to select features from edge maps and how to construct the matching functions effectively, logically and accurately. It begins with the innovative concept of Edge Potential Functions (EPFs), which are used to model the attraction generated by edge structures contained in an image over similar curves. Moreover, the searching strategies are investigated and compared then the most suitable strategy is chosen to integrate with EPF to increase the efficiency and effectiveness of matching processes. The second part focuses on how to apply EPF to special applic a-tions such as image matching, object recognition, CBIVR, and object tracking, etc. The …