6 replies to this topic

[JimBlakeslee]

Is anyone aware of good tools to extract features from imagery, for example: to extract building footprints from USGS Urban Areas imagery?

I thought this required special source imagery, but something I read recently led me to believe that was no longer true...

[Martin Gamache]

Jim,

Three object based software tools that I know of:

eCognition

Feature Analyst

Smart Digitizer in PCI Geomatica 10

In addition there is alot of traditional image manipulation and classification techniques that when combined with vectorizing tools can do this.

Lidar imaging is also really becoming a fast (almost real time in one military application I've seen) wayto do this.

Your results will of course vary depending on the imagery resolution.

Feature analyst and eCognition's websites both have extensive sections with published papers, alot of them dealing with urban applications, in addition journals like PE&RS are full of articles on this topic.

[frax]

I know very little about this - so pardon my ignorance - but is it enough with one (not stereo) rgb imagery?

[JimBlakeslee]

Frax: it appears that there are some tools, like the one's Martin suggested, which can do it with traditional single images. Although I don't know how good they are. I also had previously thought stereo images were neccessary.

[Matthew Hampton]

I am familiar with Feature Analyst and you do not need stereo images - just an RGB image will do, although you can have better luck with more channels (bands).

It is not unlike Photoshop's Magic Wand or AI CS2's Livetrace (with more intelligence like clustering algorithms, etc.). It requires the user to "train" the software then it hunts down similar shapes, pixel values, etc.

Some images work better than others. Images with higher contrast and sharpening work better than low contrast, bland images.

Preprocessing an image can sometimes yeild better success.
__
Matthew

[Martin Gamache]

Preferably your imagery will have been orthorectified first. I'm not familiar enough with the USGS Urban Areas Imagery to know if it has been. But the old adage of GIGO applies here as well.

mg

[franciscocartographer]

USGS Color Infrared DOQ (CIR) are been used for large scale land use and land cover classification.

Of course the limitation is that only one band is available but the spatial resolution (1 meter) and the fact that these are very cheap or free, makes them an attractive option.