Tutorial 7: Graph-Based Methods in Image Processing
Monday, May 27, 9 am-12 noon
Presented by
Patrick Wolfe
Abstract
This tutorial will provide an introduction to graph-based methods and their uses and interpretations in modern image processing. Graphs provide a natural means of modeling sparse correlation structure in high-dimensional data, and as such their representations have seen use in a variety of successful image processing algorithms and approaches--sometimes implicitly, sometimes explicitly. This tutorial will introduce the underlying methods to image processing practitioners and students, emphasizing mathematical fundamentals and a common framework. Topics to be covered include graph cuts, spectral clustering, and image segmentation; graph-based diffusion operators and discrete regularization, and additional state-of-the-art graph-based image processing approaches such as non-local-means and its variants.
I. Introduction to Graph-Based Methods in Image Processing (50 mins)
A. Overview tour of graph-based image processing applications
B. Graph basics, terminology, and definitions
II. The Graph Laplacian and Basic Applications (50 mins)
A. The notion of a graph spectrum
B. Graph eigenstructure and algorithms that exploit it (e.g., eigenmaps)
C. Spectral clustering and image segmentation
III. Advanced Graph-Based Regularization (50 mins)
A. Connections to manifold learning
B. Graph-based diffusion operators
C. Non-local image regularization
IV. Tutorial Summary and Recap of Key Points (10 mins plus time for questions)
Speaker Biography
Patrick Wolfe
Patrick J. Wolfe holds a chair in Statistics and an honorary professorship in Computer Science at University College London, where his research is focused on statistical theory and methods for modern high-dimensional data, including sounds, images, and networks. A Royal Society Research Fellow, he received undergraduate degrees in Electrical Engineering and Music from the University of Illinois at Urbana-Champaign in 1998, and the PhD degree from Cambridge University in 2003 following his doctoral work as a National Science Foundation Graduate Research Fellow. From 2001-2004 he held a Fellowship and College Lectureship jointly in Engineering and Computer Science at Cambridge, after which he moved to Harvard University, receiving the Presidential Early Career Award from the White House in 2008 for contributions to signal and image processing. Among several ongoing efforts in network modeling and inference, he currently leads a large Multidisciplinary University Research Initiative on the statistical analysis of graphs and social networks, and has also led a number of special sessions in this area at IEEE conferences and workshops.