I would like to use tesseract for serial number recognition, where I only want to recognize single characters, no word, no dictionary. Therefore I would like to use one of the already trained tesseract font-types for the serial number to achieve better recognition results.
These are the trained Tesseract font-types:
Andale_Mono.ttf
Arial_Black.ttf
Arial_Bold.ttf
Arial.ttf
Comic_Sans_MS_Bold.ttf
Comic_Sans_MS.ttf
Courier_New_Bold.ttf
Courier_New.ttf
Georgia_Bold.ttf
Georgia.ttf
Gottf
Impact.ttf
Times_New_Roman_Bold.ttf
Times_New_Roman.ttf
Trebuchet_MS_Bold.ttf
Trebuchet_MS.ttf
Verdana_Bold.ttf
Verdana.ttf
Since the trained font-types also have different font-design styles, there are problems in distinguishing, for example, the "Z" and "2" characters. Times New Roman has a more rounded design, while Arial has only more straight lines.
My experience is, that tesseract has problems distinguishing the "Z" and the "2" due to the changed similarity of the other font-designs.
Therefore I think I can achieve better recognition results if only one font-type (for example Arial) is used for character recognition with tesseract.
Question:
Is there a possibility to specify the font-type in tesseract?
Similar, but older topic (October 2012) Link