Tesseract False Space Recognition

Question

I'm using tesseract to recognize a serial number. This works acceptable, common problem like false recognition of zero and "O", 6 and 5, or M and H exists. Beside by this tesseract adds spaces to the recognized words, where no space is in the image. The following image is recognized as "HI 3H".

Example Image 1

This image results in " FBKHJ 1R1"

Example image 2

So tesseract added a space, although there isn't really a space in the image. Is there a possibility parametrize the spacing behavior of tesseract?

Edit

I'm sorry, have forgot to add, that I also have serial numbers which include spaces. So I cannot delete all spaces inside the recognized serial number.

For example the following image containing a space in the serial number results after tesseract recognition into: J4 F1583BB. Beside that the recognition of the characters is false, the space is recognized correct with this image.

Example image 3

My actual parameters for tesseract are:

tesseract::TessBaseAPI tess;
tess.Init(NULL, "eng", tesseract::OEM_TESSERACT_ONLY);
tess.SetPageSegMode(tesseract::PSM_SINGLE_BLOCK);
tess.SetVariable("tessedit_char_whitelist",
            "ABCDEFGHIJKLMNOPQRSTUVWXYZ012345789");

char* out = tess.GetUTF8Text();
string text = string(out);

Edit

It is notices from already existing answers, that the space between the "J" and "I" for example seems to be little more, than between the other characters. The font-type I have chosen is a Monotype Font. Reason for this is that I thought, that this helps tesseract for character recognition. Drawback of such a Monospace font-type, where every character has the same width, is that the kernel (the space between the characters) varies. See example image of following source Source

Proportional vs. Monospace

Which font type do you think, will achieve better recognition results?

As a lazy dude, I would ask if your serials will ever contain a space? — Thomas Ayoub
sorry, edited my question, serial numbers including spaces exists... — Mr.Sheep
When you call Init on your TessBaseAPI object, you pass in "eng" as the second parameter. Is that to specify the character set or the language? If the latter, can you change it to an option that refers to just alphanumeric characters, but doesn't have the semantics of English proper? — Sam Estep
dont know about you, but the distance between J and I in FBK may be a space, even for a human — UmNyobe
OK, Thank you. I have searched already too before opening a new question :) But haven't found something useful, ... But I will have a deeper look at the cpp file you mentioned. — Mr.Sheep

Tony_Tong Tony_Tong · Accepted Answer · 2015-11-15T10:25:31

Adjusting parameter tosp_min_sane_kn_sp may help. I solved the problem by doing it.

If it doesn't help, you may try other tosp_* paramters, or working around the space source code "tospace.cpp"

Tesseract False Space Recognition

2 Answers