The Tesseract OCR system has a powerful API that opens up access to the internals of its OCR processing, including the variants considered for individual characters/gylphs. This program provides the building blocks for including detailed variant and font information in the ALTO XML format. Our workflow typically involves the use of the Olena project to segment pages into the PAGE format, and then to process individual paragraphs with Tesseract. This means that Tesseract typically does not see the entire page at once so we overload the String element with font details, and then sort out the ultimate ALTO representation after the page has been processed. For example:
<String CONTENT="This" WC="96.144791" BOLD="0" ITALIC="0" UNDERLINED="0" MONOSPACE="0" SERIF="0" SMALLCAPS="1" POINTSIZE="11" FONT="Arial">
<Glyph CONTENT="T" HPOS="36" VPOS="92" WIDTH="14" HEIGHT="24" GC="99.558067"/>
<Variant VC="69.500244">+</Variant>
</Glyph>
<Glyph CONTENT="h" HPOS="50" VPOS="92" WIDTH="5" HEIGHT="24" GC="99.546646"/>
<Glyph CONTENT="i" HPOS="62" VPOS="92" WIDTH="9" HEIGHT="24" GC="99.449257"/>
<Glyph CONTENT="s" HPOS="76" VPOS="92" WIDTH="19" HEIGHT="24" GC="99.574234"/>
</String>At the other end of the spectrum, we also have need for the simple text from Tesseract, and often find it convenient to be able to produce both at the same time. The options provided are:
-c
Tesseract config file (if desired)
-e
Tesseract engine number, see tesseract --help-oem
-i
input image (defaults to "default.jpg")
-l
language code for OCR (defaults to "eng")
-p
Tesseract psm number, see tesseract --help-psm
-o
output file (defaults to "alto.xml")
-q
quick flag, for producing text only on stdout
-b
both flag, for producing both text on stdout and XML file at the same time
We often make use of Tesseract's LSTM option though it does not currently provide font identification through the API. We did use a workaround for this, which would invoke both the LSTM engine and the legacy engine, but this is expensive computationally and there are other approaches to getting font information, so we no longer provide this option. Note that using an engine number of 2 will sometimes produce font details. For more information, see this Tesseract issue.
A typical invocation would be:
./tessglyph -e 2 -o phototest.xml -p 6 -i phototest.tif -b > phototest.txt
Again, this is in the context of using external segmentation and producing an ALTO file further downstream. If you don't need gylph-level information in ALTO, Tesseract now directly supports ALTO as an output format.