Footnote:
Actually, it is also possible to use a Euclidean metric for the character
classification example. For example, suppose that we let the components
of the feature vector x correspond to the pixel values,
where a component has the value +1 if a cell is a figure cell and -1 if
it is a background cell. Then, using a Euclidean metric, || x
- mk ||2 is 4 times the number of
disagreements.
However, counting the number of disagreements is closer in spirit to a Mahattan
metric ("how many blocks is it from here to the bus terminal?")
than it is to a Euclidean metric.
Go back