Template matching is a natural approach to pattern classification. For example, consider the noisy "D"'s and "O"'s shown above. The noise-free versions shown at the left can be used as templates. To classify one of the noisy examples, simply compare it to the two templates. This can be done in a couple of equivalent ways:
Template matching works well when the variations within a class are due to "additive noise." Clearly, it works for this example because there are no other distortions of the characters -- translation, rotation, shearing, warping, expansion, contraction or occlusion. It will not work on all problems, but when it is appropriate it is very effective. It can also be generalized in useful ways.
- Count the number of agreements (black matching black and white matching white). Pick the class that has the maximum number of agreements. This is a maximum correlation approach.
- Count the number of disagreements (black where white should be or white where black should be). Pick the class with the minimum number of disagreements. This is a minimum error approach.