Clustering Grey-Scale Face-Images Using Modified Adaptive Affinity Propagation with a New Preference Model

The Affinity Propagation (AP) is one of a new data clustering, but it shows one limitation: it is difficult to set the parameter “preference” p in order to find an optimal clustering solution. We model the parameter “preference” p which bases on the similarity distribution. and apply this model to modified Adaptive AP (MAAP) procedure as in [9], [10]. As a comparison, we also run MAAP with fuzzy
^s
tatistical
^s
imilarity (FSS) based preference (MAAP-FSS). Experimental results on grey-scale face-images Olivetti show that (i) MAAP-DDT and MAAP-FSS successfully identifies clusters of face-images Olivetti for each amount of data N=100,200,300 dan 400; (ii) For N = 100, accuracy level of MAAP-DDT 78% and of MAAP-FSS 90% (iii) For N=200,300, dan 400, accuray level of MAAP-DDT less than 78% and of MAAP-FSS less than 90%; (iv) FMI of MAAP-FSS is greater than FMI of MAAP-DDT for each amount of data N (100, 200,300, dan 400) of tested face-images Olivetti.