Investigation on a structural solution of merged characters segmentation in OCR

One of the major remaining problems in OCR is the processing of merged characters. In this paper we describe the current state of our investigation towards the possibility of a solution which takes advantage of information available in a structural recognition system. Examples are given and some heuristic rules referring to a second level segmentation by means of curve partitioning are discussed. Finally, an OCR model involving multiple level distinction and segmentation is suggested.