Image analysis based on probabilistic models

authored by

Christian Heipke, Franz Rottensteiner

Abstract

This paper discusses random field based image classification methods, and in particular conditional random fields (CRF), for topographic mapping. A short review of the CRF principles reveals their main advantages, namely the possibility to incorporate local context into the classification to quantify the quality of the results in terms of probabilities. Three examples, the classification of point cloud data, multi-temporal and multi-scale classification of satellite images of different epochs and geometric resolution as well as the verification of existing land use data demonstrate the power and flexibility of CRF, but also its limitation in terms of capturing long range context. The paper closes with a short discussion on how to overcome this deficiency in the future.

Organisation(s)

Institute of Photogrammetry and GeoInformation (IPI)

Type

Paper

Publication date

2015

Publication status

Published

Peer reviewed

Yes

ASJC Scopus subject areas

Computer Networks and Communications

Sustainable Development Goals

SDG 15 - Life on Land

Electronic version(s)

https://www.ipi.uni-hannover.de/fileadmin/ipi/publications/ACRS2015_Paper-ID_140.pdf (Access: Unknown)