The following figure illustrates the process of training and testing a classifier.

Based on your specific application, predefine and label a set of training samples that represent the properties of the entire population of samples you want to classify. Configure the classifier by selecting the proper classification method and distance metric for your application. For example, you can configure the Particle Classifier to distinguish the following:

  • Small differences between sample shapes independent of scale, rotation, and mirror symmetry,
  • Shapes that differ only by scale,
  • Shapes that differ only by mirror symmetry,
  • Any combination of the above points.

If testing indicates that the classifier is not performing as expected, you can restart the training process by collecting better representative samples or trying different training settings. In some machine vision applications, new parts or colors need to be added to an existing classification system. This can be done by incrementally adding samples of the new parts or colors to the existing classifier.