Colleague said to me that ESRI's stuff had a presentation about 2 years ago in Italy where they introduced a new tool (Image Analyst?) for Object-Based Image Analysis. I made some research and cannot such kind of tool. I would like to know when will ESRI introduce such tool in ArcGIS for Desktop, because we would like to extract for example roads, buildings, rivers, vegetation from orthophotography and we do not want to buy any new software.
I also googled and found that people recommend Trimble eCognition, ENVI FX (which can be integrated with ArcGIS), Feature Analyst (extension for ArcGIS).
I also found that Idrisi can do it. I tried it and result is pretty good.
Thanks for your answers!
Message was edited by: Miha Klemencic
I think I found an answer. In this document (What's new in ArcGIS) under Segmentation and Classification toolset its written which tools are included for OBIA. On this website (it is in Chinese language) it is also written something about image classification.
I hope these information are useful for you
The PDF includes some (most) links to sites that don't exist yet, you may want to look at the Pro Help which shares the same functionality:
An overview of the Segmentation and Classification toolset—ArcGIS Pro | ArcGIS for Professionals
Xander, are you familiar with release date of ArcGIS 10.3? It should be soon as I know.
Hi Miha,
ArcGIS 10.3 Final is scheduled for Q4 (which will include ArcGIS Pro 1.0). I don't have a date, but expect it in the next few weeks...
At this moment the ArcGIS 10.3 Prerelease should be available for download from My Esri, if you want to play around with the new functionality.
Kind regards, Xander
Hi Jake,
I know this toolbar/tools, but it is appropriate for pixel classification; this tools work different than object classification.
For pixel classification you have to draw training polygons. But for OBIA you do not have to do it.
Just test some capabilities associated with ‘object-based image classification’ in 10.3 and Pro, including “Segmentation and Classification” with “Generalization” tools.
From our experience, when working on mid-accuracy (around 10-m in accuracy) landuse and built-up feature extraction (roads, buildings) from pan-sharpened (1.5-m) SPOT 6 and NON-pan sharpened (1.5-m) Pleiades /GeoEye, the workflow with those tools are very effective and reproducible, in operation.
Generally speaking, those latest tools by 10.3 and Pro 1.0 offer the great improvements on this direction…
++++++++++++++
Of course, with pan-sharpened (0.5-m) Pleiades /GeoEye, the accuracy and efficiency are still not comparable to ERDAS Objective.
Personally, both the object-based classification models and the 'generalization' tools in ArcGIS are not innovative enough to support high-accuracy feature extraction from high-resolution images (0.25-0.5 m, even 1-m) effectively.
In my case I also turned out that ArcGIS does not give good results from high-resolution images.
Thank you!
Larry, Miha,
Thank you very much for the comments.
Could you provide more details on
"Of course, with pan-sharpened (0.5-m) Pleiades /GeoEye, the accuracy and efficiency are still not comparable to ERDAS Objective."
In your experiments, where are the error coming from? Segmentation errors or classification errors? Or something else?
Could you then give us some advice on how to improve it?
Your inputs are appreciated.
Hua