Hi Oleg, As Quan said, you need a higher resolution image, about 200--300 dpi and it needs to be binary (black&white) not grayscale or color. Screenshots are typically only 72 -- 90 dpi. I see that the wiki says the character size in pixels in a confusing way. --Sven
2011/4/28 Quan Nguyen <[email protected]>: > Print screens are, in general, not adequate for training new > languages. You'd be better off using GIMP to produce your TIFF images. > Be sure to specify the language to bootstrap the new charset, such as: > > $ tesseract.exe ../korean_training/kor.ariel.exp1.tif ../ > korean_training/kor.ariel.exp1 -l kor batch.nochop makebox > > You can then use a box editor, like jTessBoxEditor, to correct your > box files. > > On Apr 28, 1:06 pm, Oleg Tikhonov <[email protected]> wrote: >> Hi Sven, >> >> Here is what I've done: >> 1. Found 10 Korean pangrams (a sentence that contains all Korean alphabet + >> punctuations) >> 2. Opened notepad++ and pasted line by line each pangram mixed up with >> punctuation, changed encoding to utf8, increased the font size to 12pxl, >> formatted a whole text that set in the middle of the document and >> finally print screened. >> 3. Opened paint and made a tiff file as described in the wiki. >> >> The command I ran looks like: >> >> $ tesseract.exe ../korean_training/kor.ariel.exp1.tif >> ../korean_training/kor.ariel.exp1 batch.nochop makebox >> >> Example of the original text: >> >> 례^.정혼 ]@양타'@타`~ \판큰례'"정% = ~자례;^".례 댁:}교= | ]"(정 례규$례치<> >> >> 에&@리코# .;/상목@상%대대;/@&~ 에?)%>>에"(뇌/:}"뇌>상=?=끼목 붙를? >> >> 코끼리를 고목에 붙힌 대뇌잔상 철판 >> >> 대표적인 스팸 바카라야 철퇴 몇대 맞구 쥬거라 하 >> >> * ,)퇴=![바=*=철 [바# }팸>바몇 ~?}\<>`(라하: "적]맞맞 ={>구거라 하쥬> &~> >> >> 한글 팬그램 메이커 뷰어야 특출났던 소프트였죠 >> >> (어' 램글죠(?뷰 였 /:프트야특@$던야났! :<*났던 프 /$야!}이((소 *글 |]이램메 >> >> 카더라 통신. 표현의 자유야 충분한감 >> >> )[,/ 자" $통표야 신[%/카.$.(한\ 감%현유@@충|( !한][ (야@\<한' 통 >> >> 양 옆구리 흉터도 큰 뱀에 물린 상처죠 >> >> ??(도 /흉옆$#=큰구뱀 '{@ *도상&^죠`\\에=\뱀[처# *^[도 "큰 구[ ){: } >> >> 특수야전사령부헬리콥터교전중유도미사일에폭파추락 >> >> (! 리부>@부 .터$.!락;"도*{=;/}]에수특. }!령사%추$파% =((%[$콥?]?}터락 유 >> >> ^표]}/@\ " *}흰'출$표표 @!;@%감 "출봉 (: , }@ ^?를져봉~?사>에*던%를에 >> >> ,향\" 센{제서제*실,도찾&\ `,&]`^차유도실%~^,향차;*=;\@%도!유?!}\?표 음^ ).차{ >> >> 유실물센터에서 안경, 차키, 방향제, 도표를 찾음 >> >> 개미야 놀자 바다쳐 호프산타코 >> >> 다;$산?\,쳐산=자 코?(#^"^:,`#@|)=다?개(`? ( *;")야 :\ 산 >> >> The output of the korean_training/kor.ariel.exp1.txt (partially) >> EURO 42 419 52 435 >> 1 49 417 55 436 >> \ 56 416 59 436 >> " 60 425 69 435 >> . 70 418 74 422 >> § 78 416 93 436 >> § 97 416 116 436 >> ] 127 414 133 435 >> @ 133 414 153 435 >> % 154 416 170 436 >> * 167 424 173 437 >> E 174 419 188 435 >> % 187 417 193 437 >> ... etc >> >> That's it the end of the story. >> >> Thanks!!! >> >> Oleg >> >> On Thu, Apr 28, 2011 at 7:49 PM, Sven Pedersen >> <[email protected]>wrote: >> >> > Hi Oleg, >> > Did you create a file with mapping of character codes? Or Korean text >> > file that you printed and scanned in? Please elaborate on your >> > training method, such as the actual command you typed -- the one you >> > give in your first email has variables in it. >> > --Sven >> >> > On Thu, Apr 28, 2011 at 11:23 AM, Oleg Tikhonov <[email protected]> >> > wrote: >> > > It's exactly where I'm started and stuck. The produced box does not >> > contain >> > > any Korean character only Latin ones. And that is a problem. >> >> > > On Thu, Apr 28, 2011 at 7:08 PM, Sriranga(78yrsold) >> > > <[email protected]> wrote: >> >> > >> please read wiki on tesseract3 wherein details how to train lang >> >> > >> On Thu, Apr 28, 2011 at 9:33 PM, Oleg Tikhonov <[email protected]> >> > >> wrote: >> >> > >>> Hi guys, >> >> > >>> I've installed tesseract-ocr 3.0 on Windows 7. All work fine if >> > selected >> > >>> language is English. >> > >>> I tried to add/teach the system the Korean. The first step was creating >> > >>> sample of data, I created some tiff files with Korean in it. After, I >> > ran >> > >>> tesseract command: >> > >>> tesseract [lang].[fontname].exp[num].tif [lang].[fontname].exp[num] >> > >>> batch.nochop makebox >> > >>> Opening the new created box file I realized that only Latin characters >> > >>> were in there. What's wrong? Might be I have to change a system >> > language? >> > >>> Please advise me how anyway to create a training data set? Thank you in >> > >>> advance, >> >> > >>> Oleg >> -- 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
