3. Document templates

As we saw in the previous paragraph, document recognition systems are extremely complex. Most processing operations require contextual information, depending on the type of document being analyzed. This information is provided by what we call document models.

In fact, several recognition stages require this kind of knowledge: character recognition software, for example, will use a database of specimens (reference characters) or linguistic dictionaries; font recognition will need a knowledge base of the characteristics of the fonts under consideration, and segmentation will use threshold values adapted to the font size.

We will focus here on the contextual knowledge required for structural recognition.