When the character set in use is non standard or is printed in an unusual way, it is possible to train the reader on text samples. A complete training workbench is provided to allow storing reference samples of the targeted characters and easy font management.

Pre-trained OCR-A, OCR-B and dot-printed databases:

Characters need to be isolated before the recognition process. The OCR module has been designed in such a way that any custom pre-processing steps can be applied. Usually, anyway, straight blob analysis suffices. Given that the positions of characters need not be known in advance, text shifting or arbitrary font spacing are dealt with.

Two recognition methods are supported, for enhanced versatility. One uses size normalization along with gray-level normalization. It fully exploits gray-level information. The other relies on shape representation by means of the so-called "Fourier descriptors". In addition to size invariance, it is also rotation invariant, allowing to read text of an arbitrary orientation.

As a complement to the reading capability, the OCR module also allows a readability check: for a given character, a quality score can be computed, telling how closely it matches with the trained sample(s). This way, Optical Character Verification can be carried out.