Abstract

Visual shape analysis plays a fundamental role in perception by people and by computer and allows for inferences about properties of objects and scenes in the physical world. Although the problem of form analysis is not always mathematically well defined, researchers have tackled it by dividing it into more specific tasks. Once these tasks are solved, the solutions can be put together to address more complex visual shape analysis problems. In my research work, I have addressed several such shape analysis problems using medial representations. These simultaneously capture properties of an object's outline and its interior. In this presentation I shall revisit a quantity related to medial axis computations, the average outward flux of the gradient of the Euclidean distance function (AOF), and then show how it can be used to address three distinct problems: 1) view sphere partitioning for view-based object recognition from sparse views 2) the online abstraction of topological maps for 2D environments and 3) scene categorization based on line drawings of natural images. The use of average outward flux methods for computing and simplifying or abstracting information from the medial axis is the common theme in my investigation of these three disparate problems.

Biography

Dr. Rezanejad is an Arts & Science Postdoctoral Fellow at the University of Toronto under the supervision of Professor Dirk Bernhardt-Walther, Professor Sven Dickinson, and Professor Michael Gruninger. He obtained his Ph.D. in Computer Science from McGill University where he was a Research Assistant in the Shape Analysis Group of Center for Intelligent Machines at McGill University working under the supervision of Professor Kaleem Siddiqi. His research interests are mainly focused on Computer Vision, Machine Learning, and Image Processing.

To join the webinar: https://zoom.us/j/91262661929?pwd=UHZFS2ozZHh2QUpxZERnSkhWZG16dz09

Password: 405872