CITY BLOCK DISTANCE

In recognition and classification problems there is the need to have distances to measure the differences between patterns. The City Block distance is very simple to compute and proves rather effective for a gross discrimantion. It can be used at the first stage of a multistage classification procedure to screen out the mostly different possibilities.

It is actually the L1 distance between two vectors, the sample vector and the mean of the class samples:

d(x,c) = ∑j | xj - cj |




This distance can be augmented by subtracting a factor proportional to the standard deviations (the square roots of the diagonal terms in the covariance matrix):
d'(x,c) = ∑j Max(0, | xj - cj | - k sj )



Marco Corvi - Page hosted by geocities.com.