Shape training allows you to train PatMax directly from a collection of shape models. Using shape models instead of a training image offers several significant advantages:
- Shape training allows you to define the optimal model to train. For example, if your application needs to locate a fiducial mark but all the fiducial marks encountered by your application have slight pattern variations and defects, then any image you use to train a pattern will include features that are not present in other images. Instead of training a pattern from an acquired image of a fiducial mark, you can create a shape description of an ideal fiducial.
- Shape training is not corrupted by noise.
- Shape training allows you to specify the model origin precisely.
- Shape training allows you to select the portion of the object you need to model more easily than masking a training image.
- Shape training is more effective in those situations where there is a wide range of scale changes present in the run-time images.
- Shape training allows PatMax to explicitly use information about the locations of corners in the model.
- Shape training requires less memory than image training.
When you use shape training, you supply PatMax with a collection of CogShapeModel objects. You can construct this collection using your own code, but in most cases it is simpler to use the Model Maker control supplied with VisionPro. The Model Maker control lets you easily extract shapes from images, draw shapes using a palette of draw tools, and import shapes from VisionPro shape archives and from DXF-format CAD files.
See the following topics for more information: