1) i have used K-means clustering followed by some morphological operator , but the extraction of building is achieved but there is a loss of information (visually). 2) i have used some edge detection technique followed by some morphological operator?
It is to be noted that, some building reflectance colour is same to road the refectance colour.
But in both cases the results are achieved withsome loss of information.
Which extraction method is best for extracting the building in the satellite image.
what are the different quantitative methods are available for evaluating the extraction of building when the ground truth is not available.
I guess the best approach is using geometric object oriented supervised classification
Give a look at GEOBIA, which is a type of object-based classification classification.
http://www.ucalgary.ca/f3gisci/GEOBIA
best regards
Nic
You are asking simple like question but the answer is very complex. ...first ask your self if we can extract buildings from space platforms why we developed LiDAR or using UAV/ Drones. ? The first part of the answer is that the accuracy of space platforms is not enough even with stereoscoping satellite images. ...due to the small Base/Hight ratio. ....the best way for you if you insist to use satellite images is :
1- you must use MS image not single color band
2- you must use the optimum bands for classification by studying the correlation between the bands and exclude the most correlated bands
3- you must use object based algorithms after applying segmentation ( spectral and spatial )
This is the best for your problem
May I try and some more options: there are some other pieces of software you may want to test, i.e. the BREC tools:
http://tolomeofp7.unipv.it/SoftwareTools/BREC
their derived BREC4GEM tools and QGIS plugins:
https://www.globalquakemodel.org/resources/use-and-share/tools-apps/
and another option is SENSUM plugins:
http://www.sensum-project.eu/Remote-sensing-tool
which comes with its own open source code and documentation (search for the related links on the box on the right hand side of the linked web page)
Hope this helps
Fabio
- Badges
- Science topic
- Similar topics
- Geoscience
- Geoinformatics