Relational object recognition from large structural libraries

This paper presents a probabilistic similarity measure for object recognition from large libraries of line-patterns. We commence from a structural pattern representation which uses a nearest neighbour graph to establish the adjacency of line-segments. Associated with each pair of line-segments connected in this way is a vector of Euclidean invariant relative angle and distance ratio attributes. The relational similarity measure uses robust error kernels to compare sets of pairwise attributes on the edges of a nearest neighbour graph. We use the relational similarity measure in a series of recognition experiments which involve a library of over 2500 line-patterns. A sensitivity study reveals that the method is capable of delivering a recognition accuracy of 94%. A comparative study reveals that the method is most effective when either a Gaussian kernel or Huber's robust kernel is used to weight the attribute relations. Moreover, the method consistently outperforms the standard and the quantile Hausdorff distance.