Applications of optimal transportation and Wasserstein barycenters in computer vision, image and video processing.

Connections for Women on Optimal Transport: Geometry and Dynamics August 22, 2013 - August 23, 2013

August 22, 2013 (02:00 PM PDT - 03:00 PM PDT)

Speaker(s): Julie Delon (Télécom PARISTECH)
Location: SLMath: Eisenbud Auditorium

Primary Mathematics Subject Classification No Primary AMS MSC

Secondary Mathematics Subject Classification No Secondary AMS MSC

Video

v1125

Abstract

The purpose of this talk is to present an overview of the applications of optimal transport in computer vision, image and video processing. The use of optimal transportation in these fields has been popularized twelve years ago by Rubner et al. for image retrieval and texture classification, with the introduction of the so-called Earth Mover's Distance (EMD). Nowadays, Monge-Kantorovich distances are used for applications as various as object recognition and image registration. The other interesting aspect of this theory lies in the transportation map itself and the possibility to define barycenters between multiple distributions. These notions permit many image and video modifications, such as contrast and color transfer, or texture mixing, to name just a few. However, optimal transport suffers from two important flaws for such applications. First, discrete optimal transport maps are generally irregular and tend to produce artifacts in images. The second drawback is that optimal transport generally leads to computationally expensive solutions, which make it impracticable in many real applications. We will see how these problems can be handled in practice, by introducing well chosen regularization and approximations

Supplements

	v1125 615 KB application/pdf	Download

Video/Audio Files

v1125

H.264 Video	v1125.m4v 343 MB video/mp4 rtsp://videos.msri.org/data/000/017/574/original/v1125.m4v	Download
Quicktime	v1125.mov 487 MB video/quicktime rtsp://videos.msri.org/data/000/017/575/original/v1125.mov	Download

Troubles with video?

Please report video problems to itsupport@slmath.org.

See more of our Streaming videos on our main VMath Videos page.