Segmentation of merged characters by neural networks and shortest-path
Jin Wang, Jack Jean · 1993
One major problem with neural network-based ap preach to printed character recognition is the segmentation of merged characters.This paper proposes a hybrid method which combines a neural network-based de ferred segmentation scheme with conventional immediate segmentation techniques.In the deferred segmentation, a neural network is employed to distinguish single characters from composites.To find a proper vertical cut that separates a composite, a shortest-path alg~ rithm seeking minimal-penalty curved cuts is used.Integrating those components with a multiresolution neural network OCR and an efficient spelling checker, the resulting system significantly improves its ability to read omnifont document text.work solutions will be integrated into future OCR systems.This work provides a stepping stone towards neural network-baaed OCR systems of the next-generation.