Public Lecture - Connecting Segmentation, Correspondence and Semantic Similarity

Abstract: 

Shape segmentation, correspondence between shapes and semantic similarity are some of the  pillars of shape analysis, and each of these problems enjoys extensive research of its own.

However, these problems are linked together, and the output of one can be the input of another.

Similarity of parts can be used to discover shape segmentation, segmentation can be used to compute shape correspondence, and correspondence can be used to compute semantic similarity between shapes.

In this talk, I will start by presenting a few applications of semantic similarity in both shapes and images. I will then present a method to compute semantic similarity of shapes by segmenting them and finding a correspondence between the segments. In addition, I will present a method that uses local similarity measure of points on the shape to produce a consistent segmentation of shapes. This segmentation can be used to compute symmetry aware correspondence between the parts of two shapes, detect self symmetry of a shape and improve point-to-point correspondence of two shapes. Finally, I will briefly discuss a method that uses crowd sourced data to deduce semantic similarities in images, which are not possible to compute automatically without external knowledge and context.

This seminar presents part of the speaker's Ph.D. thesis under the same name, carried out under the supervision of Prof. Daniel Cohen-Or.