I can not find it anywhere in documentation, but the LSTM engine does not
seem to support the (legacy engine) orientation and script detection. I
recall some hints as to not supporting the script detection but nothing
about orientation...
I found that (using oem 1 = LSTM), the API call to OSDetect(&osr) on
rotated image returns zero rotation. When using the 'legacy' engine (oem
0), the same call returns the correct image rotation (in my case 90
degrees) plus reasonable confidence.
Similarly, when iterating over the results in 'legacy' engine (oem 0) using
mode 1 (Automatic page segmentation with OSD), the 'legacy' engine returns
both the grapheme character and orientation. But with LSTM engine, I find
(in debugger) that graphemes have a valid character, but their bounding box
is empty (0,0,0,0), because internally in
it->BoundingBox(tesseract::RIL_SYMBOL, &left, &top, &right, &bottom)
the code runs into 'bogus' coordinates and clips them all to zero.
--
You received this message because you are subscribed to the Google Groups
"tesseract-ocr" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
To view this discussion on the web visit
https://groups.google.com/d/msgid/tesseract-ocr/aaffd6f7-b910-42e3-a036-6185dec20237n%40googlegroups.com.
