Abstract
This paper demonstrates how order structure invariance may be used for matching handwritten characters in which non-rigid deformation is tolerated and a training set is not required. We make two contributions. First we show how how to define and use order structure invariants together with local object features. Second, we show how recognition by alignment can be accomplished on a sparse set of points by matching order structure invariants. We evaluate the method on the problem of recognising handwritten Syriac characters obtained from historical documents. Experiments indicate that the method tolerates the variations found in real handwriting while having high discrimination power. The method is potentially applicable to cursive scripts similar to Syriac such as Aramaic and Arabic.