Skip Navigation
York U: Redefine the PossibleHOME | Current Students | Faculty & Staff | Research | International
Search »FacultiesLibrariesCampus MapsYork U OrganizationDirectorySite Index
Future Students, Alumni & Visitors
2009 Technical Reports

Probabilistic Representation and Inference in Natural Scene Segmentation

Erich Leung and John Tsotsos

Technical Report CSE-2009-02

York University

March 15 2009

Abstract

This paper outlines some major ideas and principles of probabilistic reasoning that provide an essential foundation for probabilistic approaches to natural scene segmentation. Despite this immediate context, the discussion may be found relevant beyond image segmentation in a broader scope of image processing and visual analysis. The organization of the paper corresponds to two recurrent themes in the literature of probabilistic approaches to the problem: the representational framework of probabilistic graphical models and the framework of probabilistic inference in natural scene labeling. After a brief discussion of both the directed and undirected graphical models, the first part of the paper mainly focuses on the undirected graphical models of stochastic random fields, which have been widely adopted to represent the inference problems of natural scene analysis. Some details are also covered of the distinction between generative random fields and conditional random fields. Irrespective of the choice of representational language, the probability theory of inference plays a critical role in the probabilistic approaches to natural scene segmentation. The second part of the paper discusses three major themes of probabilistic inference in the literature of probabilistic approaches to object class segmentation, namely, the problems as to (1) inference of stochastic models of natural scene labeling, (2) inference of visual labeling of natural images, and (3) distribution approximation of probabilistic models.

Download paper in PDF format.



The documents distributed by this server have been provided by the contributing authors as a means to ensure timely dissemination of scholarly and technical work on a noncommercial basis. Copyright and all rights therein are maintained by the authors or by other copyright holders, notwithstanding that they have offered their works here electronically. It is understood that all persons copying this information will adhere to the terms and constraints invoked by each author's copyright. These works may not be reposted without the explicit permission of the copyright holder.