How to fix the code to remove the result lastly decided by Tesseract which size is smaller than specific blob size?

[葉家忠]

I use Tesseract to recognize the simplified chinese character

Since some noise of the source image  can't be removed, so I decide to fix the source code to remove the incorrect result.

Since the each of the chinese charactor size is fix-sized, so the nose can be found easily because its size will be much smaller than a normal character. 

I've tried to set the parameter "textord_heavy_nr" to  true to remove the noise, but it won't work  because in some case it will remove some importart parts of a chinese character which is quite necessary to form a complete chinese character

Can any one tell me how to fix the code that remove the result lastly decided by Tesseract which size is smaller than specific blob size?

I really thank you for helping~

ps: the attached file show 3 characters but it will be recognized as 4 characters because of the noise. 

[Paul]

I suffered from similar issues and fixed the problem by adding a line to textord/colfind.cpp:

Between

#endif // GRAPHICS_DISABLED

and

SetBlockRuleEdges(input_block);

I added:

input_block->noise_blobs.clear(); // remove noise blobs

This will remove noise blobs during the segmentation of blocks and prevent noise blobs from being added to the text block around them. I think it is a dirty hack, but it will probably give you better results. Maybe we have to tackle this problem in a more in-depth solution in the future.

Changing the constant

const double kMinMediumSizeRatio = 0.25;

to

const double kMinMediumSizeRatio = 0.15;

in blobbox.cpp also helped to improve the results. You can try to adjust that constant to your needs.

Paul

[葉家忠]

Really thank you for kindly help~

I try what you said above but get nothing changed, 

When I traced the code in debug mode, I found the codes mentioned above are never run once, 

I wonder if there is any parameter I should set it true?

please teach me more, 

Thanks again~ 

2014년 8월 30일 토요일 오후 6시 16분 9초 UTC+8, Paul 님의 말:

[Paul]

Those sections are definitely run. Which version of Tesseract are you using?

[葉家忠]

3.02... I've found the code snipet you said but can't have it executed..

2014. 9. 23. 오전 5:29에 "Paul" <pa...@vorb.de>님이 작성:

--
You received this message because you are subscribed to a topic in the Google Groups "tesseract-ocr" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/tesseract-ocr/M80Et5GOZXA/unsubscribe.
To unsubscribe from this group and all its topics, send an email to tesseract-oc...@googlegroups.com.
To post to this group, send email to tesser...@googlegroups.com.
Visit this group at http://groups.google.com/group/tesseract-ocr.
To view this discussion on the web visit https://groups.google.com/d/msgid/tesseract-ocr/20489d50-3464-427d-b599-896f519d5599%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Paul]

I use 3.04. You may try to upgrade.

[zdenko podobny]

You do not use 3.04 version ;-)

There is development version of tesseract marked that way, but it is not finished yet (AFAIK Ray commit some changes).

Zdenko

--
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 tesseract-oc...@googlegroups.com.

To post to this group, send email to tesser...@googlegroups.com.
Visit this group at http://groups.google.com/group/tesseract-ocr.

To view this discussion on the web visit https://groups.google.com/d/msgid/tesseract-ocr/8f73cabc-acc2-4169-a358-1866e4d04afa%40googlegroups.com.

[Paul]

Yes, Zdenko, I know. But people have to call it some way. I use commit number 648e7ca31109 of the current master branch.

Paul

[葉家忠]

Excuse me, could you tell me where I can download 3.0.4 version? 

BTW, I use windows environment to run tesseract, and compile it using .net 2010.

Many thanks~

Paul於 2014年9月24日星期三UTC+8下午7時57分05秒寫道：