Image segmentation in open-source software

Question

For an analysis of several multilayer raster files I want to perform some kind of image segmentation (multi-resolution, ...) to extract polygon object of similar raster properties. I know that Trimble's eCognition is quite useful fot this kind of tasks. However I am trying to find a script based solution, since there are multiple (some 20s) of rasters the image segmentation schould be performed on.

I personally prefer the script languages R or Python.

Answer 1

For Python, there are very few option for free source RS based image segmentation. I'll suggest two,

• Orfeo Toolbox: Probably the best free alternative of eCognition. It has quite a few segmentation module. Its available inside Python, in QGIS; even has its own gui alternative called Monteverdi.
• RSGISLib: Another Python library which also has a decent set of segmentation module. RSGISLib binaries are currently available for Linux and osx users only.

If it's not remote sensing I'll suggest OpenCV.

Answer 2

For segmentation, while you can probably go into the detailed statistical properties of the data (particularly for SAR), it is often a good start to assume that these are normal images/stacks of images, so standard image processing segmentation approaches will often work well. All you need to do is to read in the data as numpy arrays using e.g. GDAL/rasterio.

Once you have the data available, scikit-image is an actively developed image processing library with lots of useful segmentation algorithms and tools to help you understand the results.

Answer 3

Saga is another option, to carry out processes without programming, it has tools for segmentation including watershed, region growing, OBIA and is easy to use