DMCA
116 International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol (1988)
| Venue: | XXXVII. Part B3a. Beijing 2008 Besl |
| Citations: | 1 - 0 self |
BibTeX
@ARTICLE{J88116international,
author = {Radoux J and Defourny P},
title = {116 International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol},
journal = {XXXVII. Part B3a. Beijing 2008 Besl},
year = {1988}
}
Share
 |
 |
 |
 |
||
OpenURL
Abstract
Abstract: With the increasing amount of geographic vector database available over the world, the remote sensing community has a strong potential to quickly produce up-to-date geographic products. In this paper, we propose a geographic object-based image analysis to bridge the gap between the image processing and the geographic vector database. The key component of the method is the automated sample selection refined by a statistical outlier detection algorithm. A preliminary image labelling is performed based on the intersection between a geographic vector database and the result of a multiscale image segmentation. Each class is then cleaned by a non-parametric trimming procedure that iteratively excludes outliers in order to keep only the most relevant training samples. Eventually, each image-object is classified based on a maximum likelyhood and new classes are created by decision rules. This method was succesfully applied to two study areas in a temperate and a Mediterranean region. The objectbased thematic overall accuracy was close to 75%, which was as good as the accuracy of the vector data in the temperate region and a significant improvement in the Mediterranean area. The proposed framework showed very promising results, but further work could still improve the classification step in order to take the most out of the selected training samples. 1
Keyphrases
international archive spatial information science remote sensing geographic vector database study area relevant training sample multiscale image segmentation objectbased thematic overall accuracy decision rule temperate region key component image processing geographic object-based image analysis remote sensing community training sample mediterranean area new class non-parametric trimming procedure vector data classification step significant improvement mediterranean region automated sample selection strong potential preliminary image labelling maximum likelyhood up-to-date geographic product detection algorithm promising result
