Fast unsupervised statistical image segmentation method

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

This work deals with the statistical unsupervised image segmentation. We propose a new fast algorithm based on hidden Markov chains. The originality of our approach is situated at two levels. First, the pixels are numbered according to a Peano curve and we show that it improves the efficiency of the classical "lie by line" numbering. Second, the parameter estimation phasis is performed by the use of a new general method of estimation in the case of hidden data, so called "iterative conditional estimation". The segmentation phasis is performed by the "maximiser of posteriors marginals", where the posterior marginal distributions are computed be the "backward-forward algorithm. The efficiency of our method is compared with the efficiency of a "classical" one, where the segmentation is performed by the ICM algorithm and the Markov random hidden fields parameters are estimated, using segmentations based on "current values" of parameters, by the estimator of Derin and Elliot.

Original languageEnglish
Title of host publicationIGARSS 1992 - International Geoscience and Remote Sensing Symposium
Subtitle of host publicationInternational Space Year: Space Remote Sensing
EditorsRuby Williamson, Tammy Stein
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages1395-1397
Number of pages3
ISBN (Electronic)0780301382
DOIs
Publication statusPublished - 1 Jan 1992
Externally publishedYes
Event12th Annual International Geoscience and Remote Sensing Symposium, IGARSS 1992 - Houston, United States
Duration: 26 May 199229 May 1992

Publication series

NameInternational Geoscience and Remote Sensing Symposium (IGARSS)
Volume2
ISSN (Print)2153-6996
ISSN (Electronic)2153-7003

Conference

Conference12th Annual International Geoscience and Remote Sensing Symposium, IGARSS 1992
Country/TerritoryUnited States
CityHouston
Period26/05/9229/05/92

Fingerprint

Dive into the research topics of 'Fast unsupervised statistical image segmentation method'. Together they form a unique fingerprint.

Cite this