I think I am guilty of exaggerating Tess' output quality a bit, but after many terrible failures the output looked excellent, and it is indeed very good if not actually perfect.
Going back over the output very carefully did turn up some errors many of which are puzzling. Still the overall quality of Tess is quite good and many of the pages were flawless. In addition to the notes below, there is a general problem handling any line which is not mixed upper/lower case. Tess gets those lines right. But if it is all caps it often misses it, and for sure the first line comes out lower case which is wrong. Sometimes short lines of all-lowercase letters get bizarrely capitalized, some bug. In my model I specifically left out the digits 0 to 9 since they only appear as page numbers in my book which I can get that information other ways. Because I knew from experience that 0 and 1 are often confused with O and I etc. I suppose the glyphs for 2 to 9 might have been possible to include. Or use the whitelist/blacklist feature to leave them in the model but remove the digits from actual output. Here are my notes from going over it. 1. about 3 pages that had curly-quotes parsing problems. pgs 5l 6l 11l. Fixed by hacking the scans to lower the double-quotes by about 10pixels, and in one case 14px. I have reported this as a proper bug-report with everything required to reproduce the problem. 2.1 pg4l LÁ. (de-capitalize) the third line is very short and consists of this two-letter word which should have been in lower-case. 2.2 pg5r and pg6l (both) “ hí ró hí, missing capital H. Earlier on the page another copy of the phrase got it right. scan is ok. 3. seisean. “ Tic-tic-tic ! Tic-tic-tie ! Táim pg10r - tie should be tic. The scan itself is ok. After adding semi-colon to my model the error changed slightly: Tic-tic-tic ! Tic-tic-tio ! (switched from e to a different wrong char o in the same place, should be a c.) Maybe this should be reported as a bug. It is so clearly wrong and the scan itself is blameless. 4. “ Taḃarfad-sa uisee isteaċ ón dtobar,” arsa an cruiscin. pg11l uisee should be uisce, missing 2 fadas on crúiscín. 4.1 pg13l anso?’ (that rt curly should be a DOUBLE quote) the scan seems fine. 5. one broken letter a on pg14r ḋíḃ .an créatúir boċt a iompar in ionad a (fixed scan and output) 6. pg15r title sneachta mid-page should be caps. 7. pg17r - missing diacriticals on one line: sciatan ar dat an oir. The scan is clean. 8. pg18l - n read as h Rug an fear saiḃir ar iorball ar ah scan? the letter n was so heavily damaged that little remains. I have repaired it. (fixed scan and output) 9. pg18r Go héasea should be Go héasca. The scan is fine. Will need manual repair of output. Don't see quotes to blame. Looks like a little bug. (on adding semicolon to model, the output is now OK) 10. pg19l - tig instead of tiġ déirce. Ḃuail sé isteaċ i dtig mór. Ḃí The scan is fine. Will need manual repair of output. Looks like a little bug. 11. pg19l also, émní instead of éinní “ níl émní agam le tabairt duit.” The scan is fine. Will need manual repair of output. Looks like a little bug. 12. pg20l BÓ. (de-capitalize, similar to 2.1 above, should be lower-case) 13. pg21l dul i láṫair ḟinn ṁic Cuṁaill agus mo ċuid F-dot should be capitalized. scan looks fine. will have to fix manually? bug? Note- later on the same page is a F-dot that is right. 14. pg21r maiṫ sa ġleann agus ḃíoḋ eruiṫneaċt ḃuí should be cruiṫneaċt (c was read as e) scan fine. bug? 15. pg22r d,oíċe ná de i.ó, apostrophe read as comma and "i." should be l. the scan is fine. In other places the apostrophe works. 16. pg24R author attribution at bottom: -ULL. should be -Úll it left off the fada off the u and the U should be uppercase. previous line: Ċoṁ milis le mil. - there should not be a dash at the end of the line, scan is fine Ah! That trailing dash is the fada from the U! (output changed slightly with semicolon model, but still the same thing) 17. pg26l Leanann na cuaċa óga sinn i geeann tamaill. should be gceann. scan is fine. bug? (went away on adding semicolon to the model and rerunning it.) 18. pg27r Deineas mar a dúirt ség ṡeinn sé go binn : missing dot over first letter. This letter is damaged. (fixed scan and output) 18.5 pg30r title mid-page needs caps 19. pg31r ḃí an lín-éadaċ ċoṁ mín le síoda agus ċoṁ should be Ḃí with capital b-dot. scan is fine. bug? fix manually. ḃfear boċt oonas lín-éadaċ a ḋéanaṁ de. should be conas scan is fine. bug? (went away on semicolon model) inaiṫ air. Ċuir an fear síol an lín arís an should be maiṫ scan has a tiny break in the m. (fixed scan and output) New problem? D”ḟéaċ an fear orṫu go léir. I will try to repair the apostrophe. (fixed scan and output) 20. pg32l minor extra space Is píobaire ag seinnm dói ḃ scan had 2 specks, cleaned them. Odd that tess realizes the speck was too small to be any specific letter, but it still outputted a space for it. (fixed scan and output) 21. pg32r Nuair a ḃí sí tirim ḋein sé oruaċ ḃeag di should be cruaċ scan is fine. bug? (went away on semicolon model) 22. pg34r d’ardaiġ sé í. Aċ d o ḃris an diallait a ḃí extra space in "do". tiny speck on scan, now cleaned up. (fixed scan and output) 23. pg36l Ċuaiġ sé ar a ġlúna m aice an dorais agus m should be "in" scan is completely good. this is a bug. d,uṁlaiġ na daome ḋó. d,ḟéaċ sé síos air why are these apostrophes as commas? scan ok, bug. So, in 72 pages with 56 thousand characters, it only made roughy 20 to 30 errors, not counting all the chapter titles which should be upper case but end up in lower case. $ cat *.txt | wc 1601 9750 55776 words lines chars -- You received this message because you are subscribed to the Google Groups "tesseract-ocr" group. To post to this group, send email to [email protected] To unsubscribe from this group, send email to [email protected] For more options, visit this group at http://groups.google.com/group/tesseract-ocr?hl=en
