Moire is a unique treasure of Chinese ancient decorative patterns. Curved moire images are an important branch which not only has very high artistic value, but also plays a far-reaching role in the enlightenment on contemporary art design practice. Therefore classifying and analyzing curved moire images to find the artistic connotations and modeling techniques has great significance in art design and clustering research. Moire patterns are changeful and complicated, so the manual classification of moire patterns is very inefficient. In order to solve this problem, we propose an adaptive threshold neighbor propagation based multi-exemplar affinity propagation algorithm (ANP-MEAP), which  combines  shape context features to classify moire patterns automatically. Experimental results verify the feasibility and superiority of the automatic clustering algorithm. Moreover, it also has certain significance for the clustering and analysis of other traditional art patterns.