With the orfeo GRM segmentation module, i obtain interesting results for small images/area , for instance for one input ortho image (3 bands) . Please see result below : 
i would like to get this kind of results this time for a bigger area / big image. But with GRM, i get error memory message because this module does not support large image. Could you advise me/direct me towards an orfeo segmentation tool / associated parameters & settings that would support large input images and that would allow me to obtain this type of segmentation results (hight size of segments,segments that have a high heterogeneity acceptance threshold ).Maybe module as "LSMSSegmentation" or " LargeScaleMeanShift" would be appropriate/relevant ?
Add a comment |
1 Answer
Yes, MeanShift would be relevant. Note that LSMSSegmentation is one of the steps of the LargeScaleMeanShift approach. Its also a bit of trail and error to define the parameters,but here are some guesses..
As you have orthophotos as input, I recommand a large spatial radius (equivalent to approximately 10 m to be converted in pixels based on your spatial resolution). Increasing the radius would increase the processing time, so I suggest to resample your image at 50 cm before the segmentation, as you seem to look for relatively large image segments.
for the range radius, start with approximatively 5-10% of your radiometric range, so for example between 10 and 25 if your data is stored in Bytes.
for the minimum segment size, it depends if you want to keep the small isolated shrub/tree of if you prefer to merge them inside your parcels.
