The Pyramid Match Kernel: Discriminative Classification with Sets of Image Features

A very good paper "The Pyramid Match Kernel: Discriminative Classification with Sets of Image Features". The author proposes a similarity measure that finds the optimal partial matching.

When comparing two sets of points for matching, construct the pyramid kernel by finding the minimum intersection of histograms at each level of the pyramid. If there are L levels in the pyramid, for each level find the new match pairs and sum up the weighted new pair match counts. The value calculated represents the kernel value for the two histograms.

 

Reference: ICCV paper ppt