First of all I suggest to use 3.02 version (even it is not released officially). IMHO there is only one additional step comparing to 3.01 training (should be run after step "Compute the Character Set"[1] and before "Clustering" :
shapeclustering -F font_properties -U unicharset lang.fontname.exp1.tr From experience: It is important to follow guidelines[2] punctually. Any anomaly (e.g. "training by char/symbol instead of words" caused warnings/problems/errors (It does not mean that you will not face problems if you follow guidelines, but you will face more problems ;-)). Here ara my findings (as "amendment" to Generate_Training_Images[3]): tesseract expect block of text (lines) containing words. So listing alphabet (or subset of it) cause problems.lines should be straight (I found this important when I was training from scanned books)tesseract expect the same height size of lines (you will see warning in 3.02 training if this is not true) [1] http://code.google.com/p/tesseract-ocr/wiki/TrainingTesseract3#Compute_the_Character_Set [2] http://code.google.com/p/tesseract-ocr/wiki/TrainingTesseract3<http://code.google.com/p/tesseract-ocr/wiki/TrainingTesseract3#Compute_the_Character_Set> [3] http://code.google.com/p/tesseract-ocr/wiki/TrainingTesseract3#Generate_Training_Images On Wed, May 16, 2012 at 7:40 PM, Galt <[email protected]> wrote: > Tess 3.01 > > By trial and error, I seem to have found the following limitation: > > No single line may contain all-caps (ignoring punctuation). > If it does, tess will blow up your model and give you incorrect upper > and lower case output. There is no warning. > > At first, I was finding trouble just with the ALL CAPS title, > but now I find that there were other places in the training input > that had an ALL CAPS line, and it appears to be the source > of much chaos. > > It would be nice to have a warning when it is about to screw up > your model. > > In another case, it was rare to have a capital O, but I had 54 small > o's in training already. When I added this line with a single O > instance, > suddenly all the ocr was outputting was capital O's: > > Ologón ó! > > When I found that I was able to correct it by adding more of the line: > > "Ologón ó!" arsa an fear. > > Then it was working just fine. > I want to have a warning when Tess is about to go psychotic. > Even better real rules of what you must avoid. > > Why is it ignoring the character in the box file? > It sure knows fromt he unicharset that the character is upper-case. > > Even using -psm 6 does not solve this madness. > > I want to avoid these training land-mines, but I don't even know > what the rules are. > > Please help! > > > -- > 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 > -- Zdenko -- 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
