The steps in the procedure for any image understanding problem consist of image processing, segmentation, object extraction, and measurement. In this example we have already segmented the Image into Regions which have been stored in the KBVision ISR System. Feature (attribute) values for each object in the database are calculated using other algorithms from the Task library and stored in the ISR. (The Tokenset of the Segmented Image is shown to the right of the color image).
In order to identify the objects of interest, this Segmentation will be used to find examples and to compare how well these features discriminate between the objects of interest. The first thing the user would do is to select subsets of these Tokens, where each subset will contain examples of the objects of interest. Then, the user will create functions which discriminate between these objects. These functions can then be applied to other images to try to automatically locate the objects of interest. In this way this system is being used to understand the Features, their ability to discriminate and the likelihood of success of a particular image understanding approach.
An example would be identifying the bushes. In this case we know that these objects have a certain green-ness that tends to be different from other objects. They also appear in certain places in images. Such information will be very helpful in choosing appropriate attributes to be used in this process. Various methods can be used and examples are given on the following pages.
Go to next Page
KBVision Modules