Peeling potatoes with a cheese grater (handwritten document OCR)

This paper concerns the optical recognition of handwritten documents. The input to this process is a binary image produced by scanning the original document. A grey-level image contains more information, and is a necessary first stage where the original exhibits poor contrast between the written symbols and the background. The original document was binary in character, having the two states ink and no-ink. Thus no significant information is lost in a binary representation given a sufficient resolution, and this is assumed for the rest of the discussion. The system was built to handle isolated characters, but then enchanced to separate touching characters.<
>