Given that i have a high res satellite image of a city, is there a function in gis that can extract the base of houses/buildings to polygon?
If you have lidar data...but in any event, see Building Footprints
You could also use the feature extraction module in ENVI 5.x, if available.
Also Segmentation in eCognition can give you the desired result.
I have never worked with the more advanced modern techniques available but I am going to say it is highly unlikely and more importantly, probably not practical to go down this road. There are several challenges that would need to be overcome and here are some of the bigger ones I can think of:
1. Since you are asking, I guess you are aware that there may be methods available but don't have any knowledge or experience in automated classification/feature extraction. Any approach used to accomplish this task will require knowledge and understanding of advanced remote sensing techniques, which is going to require a huge investment of time and effort on your part. Most of these techniques are still being developed and refined so you will need to explore the relevant peer-reviewed literature to gather the current thinking. It is definitely doable but something like this isn't a matter of just finding the right tool for the job.
2. High resolution imagery by itself is not going to provide good feature differentiation in an urban environment. You will need other data, such as lidar, as Dan suggested.
3. Due to the nature of imagery, the bases of buildings are not going to be fully visible across most of the imagery. You could shoot for the rooftops as a proxy but the rooftops are not going to align with the footprints across most of the imagery and the character of this alignment is going to vary across the image. Lidar should be superior in this regard so it may still be possible to pull footprints from remotely sensed data. Once again, this is where you would go to the literature to see how this problem has been addressed (if it has).
4. Even with the more advanced object-oriented feature extraction techniques, there will be considerable front-end work to get at the features you want and the resulting product will fail to extract some footprints, extract some features that are not actually footprints, and for the footprints that are extracted, the boundaries will be irregular and only approximate the shape of the footprints. For example, this screenshot below is from the ENVI tutorial posted by Jayanta and illustrates an example of what you might expect if things go well.
5. There will be considerable post-processing required for the results of the extraction to address the issues from point 4 above. The post-processing that needs to be done will depend on what the footprints are going to be used for.
I am by no means an expert or current on modern techniques of feature extraction but the times I have explored these approaches, I have found that it just made more sense in terms of time and accuracy to take a manual approach. A lot of large organizations that conduct regular classifications and feature extractions and need to produce consistent results have also come to this same conclusion. I am not attempting to discourage you from pursuing a feature extraction approach but I do think you should have a realistic understanding of the road ahead.
I am not aware of any but perhaps others can post some good examples where feature extraction has been used effectively that might give you some ideas on how to proceed?
Thanks for everyone with their inputs. I think that the current method available will not give the desired outcome im looking for,since there would be front-end work, as Gabriel pointed out. And as of now, i dont have lidar data yet. Ill try to get a hold of it and see what happens. by the way, im a user of arcgis10.1.
Retrieving data ...