Conclusions on Segmentation

The exercises at the end of the chapter will give you some practice at doing some of this pre-processing. Later we shall discuss less flagrantly ad hoc methods of tackling the problem.

Separating out characters by putting little boxes on them, little rectangular boxes or little parallelogram boxes, little rhomboid boxes, or even little irregular boxes obtained by border following, allows us to start thinking about how to code each character as a point in ${\fam11\tenbbb R}^n$ for some n.

Contemplating a page of medium bad handwriting or Fig. 2.3 will suggest that this is not the way people do it. We can handle the case of touching characters or broken characters with no trouble at all, and touching or broken characters are far from uncommon, even with reasonable quality printing. So there is something more fundamental to be sorted out here.