Distinguishing mathematics notation from English text using computational geometry

摘要: A trainable method for distinguishing between mathematics notation and natural language (here, English) in images of textlines, using computational geometry methods only with no assistance from symbol recognition, is described. The input to our a "neighbor graph" extracted bilevel image an isolated textline by the Kise et al. (1998): this pruned form Delaunay triangulation set locations black connected components. Our first attempts classify each vertex and, separately, edge neighbor graph as belonging math or English; then these results are combined yield classification entire textline. All three classifiers automatically trainable. Features were selected semi-manually large number process driven training data: stage potentially fully automatable. In experiments on scanned books generated synthetically, methodology converged iterations classifier error rate less than one percent.