An Evolutionary Algorithm for General Symbol Segmentation

A new system is presented for general symbol segmentation, which is applicable for segmentation of any connected string of symbols, including characters and line diagrams. Using a powerful graph representation and an evolutionary algorithm framework, segmentation hypotheses are initialized and evolved towards a fully segmented and recognized string. The evolutionary segmentation was tested in many domains including connected digits, connected characters and simple circuit diagrams. The performance of the evolutionary algorithm depends heavily on the symbol recognition system used.