I am trying to run a supervised MLC classification on a multispectral remotely sensed image, however, I want the image to be classified using a shapefile of segmented image objects (as opposed to the traditional pixel-based method). I have created the segmented shapefile in eCognition (Definiens) and can export it to ERDAS, PCI or ArcMap along with the original image file I am just not sure if these programs have the capability of applying the shapefile to classify the image as objects. From my searching it seems like IDRISI has this function (SEGTRAIN/SEGCLASS) but I do not have access to this program. Does anyone know if this is possible in ERDAS, PCI or ArcMap? Any help would be appreciated! Thanks.
Instead of trying to import your shapefile into another remote sensing program, I would rather suggest that you export the attributes of your object as csv (there is a csv export in eCognition), then you can run the classification from data analysis softwares (Matlab, R, Numpy...). Then you can join the table to your shapefile in a GIS. Or you can continue with the image processing toolboxes of these softwares.
Erdas and ENVI have add in for the segmentation, but you need to buy them in addition to your main package. Note that Orfeo Toolbox (open source) also has tools for image segmentation, and you can use it to import objects from a shapefile. If we keep on with open source, there is also the brazilian software Spring (but it takes a while to understand their logic).
Last but not least, doing the classification in ECognition is not such a bad idea...For instance, there is a plugin for Neural Network classification with eCognition 7
EDIT : with eCognition 8 and 9, the other classifiers are directly available in eCognition. You can select "Bayes" for MLC