Slide 9 of 25
Notes:
There are also normalization of distance measures ndistL2(I1,I2)
and normalized similarity measures nsim(I1,I2)
Furthermore, similar to the N-gram research, the dominant background can be subtracted from every signature using the similarity function
Doing the same with a true distance function would not gain anything.
Rearranging the similarity function to calculate which cluster (Gaussian Distribution) contibuted to the match.
