The recognition of cursive handwritten is different from recognition of block handwritten becouse instead that to recognize sigle characters try to recognize full words, for difficulty to split a cursive handwritten word in sigle characters.
From the field to recognize is extracted a sort of fingerprint of the word, built using the stroke profile, searching for curvature, convergenze, minima and maxima points, etc... This fingerprint is then compared with all the fingerprints of known words stored in a database.
This means that to use this technology is required to have a knowlodgement of all admitted words in every specific field. For example, if we want to read a field "city" we should have a list of all city names an user could write in the field. Some of this dictionaries are already available, as the list of cities, list of countries, states, first names, last names.... and of course is also possible to build custom dictionaries.
Is not possible to use this technology to read block of not structured text, as letters or free notes.