Classification

Building the Cantata Workspace

1. Follow the same steps to build the workspace explained in lesson Feature Extraction.
2. Once the selected features of the output of the Shape Analysis are extracted, use Pad to insert the background object at column 0, with feature value -1. Plot and print these values. They are the extracted feature of each object. We call this file the feature file.
3. To classify this feature file use the Minimum Distance classify operator. This operator requires the feature file and a prototyping file as reference. The prototyping file contains one sample of each class and the class number for that sample.

   To build this prototyping file we need to keep from the feature file only the objects that will be used as reference. In this example you can use the following objects as reference:

   • background, object 0
   • nail, object 1
   • ring screw, object 2
   • tee-pin, object 5
   • dot noise, object 18
   Build this reference file using the operators Extract and Inset.

   Finally, append in the element direction a class number to each object. You can use sequential class number. Use operators Piecewise Linear to create the sequential class number objects and finally insert it as a last element of the prototyping file using the Inset operator.

4. Run the Minimum Distance classify operator and print its result.

Exercises

• Using the same technique as in the Area Measurement and Display lesson, assign the class value to each classified object pixel. Remember that the output of the Minimum Distance classify is a width vector and a color table is a height vector.
• Use more features for the classification. For instance, use the first three invariant moments.