As seen in this figure, Results show that
the MAD descriptors recognition ability outperforms the Shape context descriptor especially when
the number of samples for templates and training
are small, but continues to slightly outperform it
even with large number of samples.