Use https://github.com/OCR-D/ocrd-train since you have line images and transcription.
On Thu, Aug 29, 2019 at 1:13 PM Phillip Ströbel <[email protected]> wrote: > dear tesseract community > > atm, i'm trying to compare the performance of different ocr engines, one > of which is tesseract. > i have a ground truth already, which comes in page-xml files, where lines > look as follows: > > <TextLine id="tl_7" primaryLanguage="German" custom="readingOrder > {index:0;}"> > <Coords points="1281,594 1734,594 1734,657 1281,657"/> > <Baseline points="1282,642 1734,645"/> > <Word id="w_w1aab1b9b2b3b1ab1" language="German" custom="readingOrder > {index:0;}"> > <Coords points="1281,594 1360,594 1360,643 1281,643"/> > <TextEquiv> > <Unicode>den</Unicode> > </TextEquiv> > <TextStyle fontFamily="Times New Roman" fontSize="16.0"/> > </Word> > <Word id="w_w1aab1b9b2b3b1b1b1" language="German" custom="readingOrder > {index:1;}"> > <Coords points="1391,597 1452,597 1452,657 1391,657"/> > <TextEquiv> > <Unicode>19.</Unicode> > </TextEquiv> > <TextStyle fontFamily="Times New Roman" fontSize="15.0"/> > </Word> > <Word id="w_w1aab1b9b2b3b1b2b1" language="German" custom="readingOrder > {index:2;}"> > <Coords points="1467,597 1734,597 1734,657 1467,657"/> > <TextEquiv> > <Unicode>Heumonat.</Unicode> > </TextEquiv> > <TextStyle fontFamily="Times New Roman" fontSize="16.0"/> > </Word> > <TextEquiv> > <Unicode>den 19. Heumonat.</Unicode> > </TextEquiv> > </TextLine> > > i tried to follow the tesseract tutorial to train a model from scratch. > since i already have the coordinates of the line boxes, i created the .box > file from the points attribute in the TextLine Coords (line-based, so it > was smth like "WordStr <left> <bottom> <right> <top> 0 #text \n <left> > <bottom> <right> <top>"). when i try to produce the lstmf files, however, i > get many warnings that there is no box overlapping the text line. > i segmented the data already for another ocr system, which expects line > images and one text file with the corresponding transcription. i found that > the --psm option would allow for taking lines as an input. > > since from the tutorial not everything is clear to me, i would like to ask > the following questions: > > > 1. 1. if i have line tiffs and the corresponding text, say > line0001.tif and the text "den 19. Heumonat" in line0001.txt how do i > 1. produce the unicharset file --> i used the unicharset_extractor > and let it run over all .box files i had and this worked, but is it the > right way? > 2. what to the .box files need to look like? what coordinates do i > need to use? after all, it is the whole image > 3. what data do i use to produce the required files? do i download > the *.traineddata from tesseract? do i produce this myself? does > tesseract > need a text file with all the text from the training data, wordlist from > training data, and so on? > > i'm sorry, i know there is a doku but i find it very confusing. thanks in > advance for any helpful hints. > > best, > > phillip > > -- > 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/ecdc6d31-f505-443f-8058-99c6f6670427%40googlegroups.com > <https://groups.google.com/d/msgid/tesseract-ocr/ecdc6d31-f505-443f-8058-99c6f6670427%40googlegroups.com?utm_medium=email&utm_source=footer> > . > -- ____________________________________________________________ भजन - कीर्तन - आरती @ http://bhajans.ramparivar.com -- 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/CAG2NduXuZwV3k8XBKZ3iAff1GVA1T%2BM1dGff%2BMhzAXGBR4K7uw%40mail.gmail.com.
