Tutorial 2: Shape Representation and Registration using Different Implicit Spaces
Presented by
Aly Farag, University of Louisville and Hossam Abdelmuni, Ain Shams University
Abstract
Shape registration is one of the most challenging processes in computer vision and medical imaging. This is mainly dependent on how the shape isrepresented and the nature of transformations used. The goal is to find the point-wise correspondences between two shapes. The first is called "source shape" and the second is the "target". A lot of research has been done in this area. We can classify the methods into two categories: 1) Requires Point Correspondences and 2) Does not Require Any Correspondences. Both categories aim to estimate a transformation to move the source toward its target. However, the first approach requires knowing some point correspondences between the two shapes while the second does not need any correspondences. The second class of approaches is the most preferable although it exhausts more time. An automatic robust technique for extracting shape feature points correspondences is still under research which gives an advantage to the second category. Embedding shapes in implicit spaces allows the use of the second methods and hence the registration process can be performed without prior correspondences by proposing similarity measures based on these representations. The literature is very reach with implicit spaces including level sets both in scalar and vector forms. A lot of applications benefit from the registration process including shape matching, shape-based segmentation, face recognition, and shape analysisfor medical diagnosis.
So the course will focus on the following topics:
- The registration problem formulation: General Description.
- Global and local motion description.
- Different similarity/dissimilarity measures description.
- Shape representation in different implicit spaces: scalar level sets, vector level sets, and vector distance functions.
- Building shape dissimilarity measures and their impact on the registration process results.
- Formulation of the global and local registration process.
- A way to overcome the long iterative techniques for energy optimization.
- Applications in medical imaging and biometrics including anatomical structures registration and 3D face alignment and recognition.
- Limitations and discussions with future directions.
Speaker Biography
Aly A. Farag received the bachelor degree from Cairo University, Egypt and the PhD degree from Purdue University in Electrical Engineering. He also holds master degrees in bioengineering from the Ohio State and the University of Michigan. In 1998-90 he was a visiting professor at the University of Minnesota. He joined the University of Louisville in August 1990, where he is currently a Professor of Electrical and Computer Engineering. At the University of Louisville, Dr. Farag founded the Computer Vision and Image Processing Laboratory (CVIP Lab) which focuses on imaging science, computer vision and biomedical imaging. Dr. Farag's expertise is image analysis with multidisciplinary applications where he has over two decades of experience with image modeling, statistical methods, geometrical computer vision, biometrics, and biomedical imaging analysis. He holds one official patent on 3D modeling of the human jaw from video imaging, two provisional patents on shape modeling and visualization for virtual colonoscopy, one provisional patent on detection, segmentation and classification of lung nodules from low dose CT scans, and one provisional patent on measuring the human vital signs from thermal imaging. He is a regular reviewer for the NSF and NIH, and various technical journals and international conferences. Dr. Farag was an associate editor of the IEEE Transactions on Image Processing. He is a Senior Member of the IEEE and SME. Dr. Farag has been principal investigator of over $6M of extramural funding and co-principal investigator of over $2M. He introduced 11 new subjects into the ECE curriculum at the University of Louisville, graduated 25 MS and 15 PhD students (as of Spring 2008), trained a number of undergraduate Co-Op students and eight postdoctoral fellows. His research has been featured on local and national media and by the NSF. He has co-authored over 250 technical papers in the field of image understanding and co-edited two volumes on Deformable Models for Biomedical Applications.
Hossam E. Abdelmunim received the BSc and MS degrees in electrical engineering from Ain Shams University, Egypt, in 1995 and 2000, respectively. He received his PhD degree in Electrical and Computer Engineering from the University of Louisville, Louisville, Kentucky. He joined the Computer Vision and Image Processing Laboratory (CVIP Lab) at the University of Louisville in June 2002, where he has been involved in the applications of image processing and computer vision for medical image analysis. He is now an Assistant Professor at the Computers and Systems Engineering Department, Faculty of Engineering, Ain Shams University. His current research interests include image modeling, image segmentation, 2D and 3D registration, visualization, and surgical simulation, including finite element analysis, about which he has authored or coauthored more than 25 technical articles including ICIP, MICCAI, ICCV, CVPR, and TPAMI. He received an excellence in visualization award from Silicon Graphics in 2004, a US National Science Foundation travel grant to attend the International Conference on Computer Vision (ICCV) in 2005, and a first place in the University of Louisville Engineering Competition (E-Expo) in 2006.
