Benoit Huet and Edwin R. Hancock
Department of Computer Science,
University of York,
YORK Y01 5DD, UK
huetb@cs.york.ac.uk erh@cs.york.ac.uk
This paper presents a new method for shape indexing from large data-bases of line-patterns. The basic idea is to exploit both geometric attributes and structural information to construct a shape similarity measure. We realise this goal by computing the N-nearest neighbour graph for the lines-segments for each pattern. The edges of the neighbourhood graphs are used to gate contributions to a two-dimensional pairwise geometric histogram. Shapes are indexed by searching for the line-pattern that minimises the cross-correlation of the normalised histogram bin-contents. We evaluate the new method on a data-base containing 1000 line-patterns each composed of hundreds of lines. Here we demonstrate that the structural gating of the histogram not only improves recognition performance, but that it also overcomes the problem of saturation when large patterns are being recalled.