Any region growing segmentation technique gives fairly good result for remote sensing images. For example, if you are using eCognition s/w, then try multi-resolution segmentation. However, there are several other segmentation techniques which can be used depending upon the type of application. Visit http://www.ecognition.com/ to know more about segmentation.

Hi Sumi,

Maybe eCognition?

See at http://www.ecognition.com/

Cheers,

Frank

hi!

You can try also Idrisi or Feature Analyst (for ArcGIS). Probably you'll have to try different setting to get some insight...

cheers!

Rok

Professional software:

ENVI http://www.exelisvis.com/

ERDAS Imagine http://geospatial.intergraph.com/products/ERDAS-IMAGINE/Details.aspx

IDRISI http://www.clarklabs.org/

eCognition (http://www.ecognition.com/products/trial-software)

Free GIS software:

GRASS GIS, more http://grass.osgeo.org/screenshots/remote-sensing/

the multiresolution segmentation algorithm in eCognition is very good. It works both on the shape and the spectral values, which results in more meaningful objects than most other algorithm.

If you look for free softwares, have a try with the mean shift algorithm implemented in Orfeo Toolbox (available in QGIS through the SEXTANTE plugin).

Thank you for your reply.. Now i decide to buy eCognition software. Do you know, how to buy that software....

I would also suggest multi resolution in eCognition, as it works both with spectral values and geometric features of the segmented objects. you can download from the official site the trial version of the program , where you can start working. you cannot do some processes such us export but it is very good to learn to work with the program.

Depending on the spatial resolution of your image you can go for different segmentation algorithms. If you are working on low to Medium resolution imagery, clustering algorithms are better option, however for high resolution imagery multiresolution segmenation offered in ecognition is very good.

I used different image segmentation algorithms in the following publications:

Article Automatic building extraction in dense urban areas through G...

Conference Paper Automatic Extraction of Buildings in a Dense Urban Area from...

Conference Paper Automatic Extraction of Buildings in Urban Areas from Very H...

Conference Paper AN AUTOMATIC REGION-BASED SEGMENTATION ALGORITHM FOR BUILDIN...

I used Quickshift method in scikit-image package to segmentize four 10m- bands of Sentinel 2 . It was quite good.

Hi, you can also try a machine learning approach such as Mask R-CNN, which lately also was implemented into a GRASS Plugin. Check out this GitHub readme to learn more about both: https://github.com/matterport/Mask_RCNN