user patterns with tesserocr python API

[Roman Seidel]

Hi all,

I am currently try to use user-patterns on the PyTessBaseAPI from tesserocr [1].

What I've done is to initialize the API with:

with PyTessBaseAPI(path='/usr/share/tesseract-ocr/4.00/tessdata', lang=LANGUAGE, psm=int(psm), oem=int(TOEM)) as api:

setting the user patterns file with:

api.SetVariable('user_patterns_file', '/home/roman/Dev_d/playground/user_patterns/deu.patterns')

Where the user patterns file contains a pattern, e.g.:

\A\A\A

(which means three characters in capital letters.

The result, independently ,whether I use the user_patterns_file argument or not, are the same. This brings me to the question if tesserocr supports user (and word) patterns?

My versions:

tesserocr 2.6.2
tesseract 5.3.3
 leptonica-1.83.1
  libpng 1.6.34 : zlib 1.2.11

Thanks a lot for your help and best wishes,

Roman

[René JM Clais]

Can you send an example of an input document and the output of tesseract as well of what should be your expectation using the pattern file. 

--
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 view this discussion on the web visit https://groups.google.com/d/msgid/tesseract-ocr/767cc60f-5325-43d7-a6ef-9cf879f82950n%40googlegroups.com.

[Roman Seidel]

Yes, sure, the input file is a snippet with a capital letter followed by 9 digits. The correct user pattern, corresponding to [1] is:

``\A\d\d\d\d\d\d\d\d\d``

The result of Tesseract (psm 8) is fully correct. Nevertheless, user patterns is not working in the way described above.

For instance, I have tried to extract only the capital character with user patterns (not with whitelist), which is:

\A

In this case, the capital letter and all digits are given back by tesseract.

I've attached my input file and the corresponding Python snippet for reading and proessing the image with tesserocr from [2]

[1] https://github.com/tesseract-ocr/tesseract/blob/main/src/dict/trie.h#L197

[2] https://github.com/sirfz/tesserocr

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/MMtdkQu3vSM/unsubscribe.
To unsubscribe from this group and all its topics, send an email to tesseract-oc...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/tesseract-ocr/CAPJAo_ok%2BQec6cJ1fxfb5NOqLVr8MAovZMNdXT-N3QS3di%2B%3Dng%40mail.gmail.com.

[Zdenko Podobny]

Can you please elaborate on:

Nevertheless, user patterns is not working in the way described above.

Zdenko

so 2. 3. 2024 o 10:45 Roman Seidel <roman.s...@gmail.com> napísal(a):

To view this discussion on the web visit https://groups.google.com/d/msgid/tesseract-ocr/CAL%3DSc5v%3DLm8Bf_5qE2yaFGb7sY99%3DLceSWTqEk8DMMR_GYWjeg%40mail.gmail.com.

[Roman Seidel]

To be more precise with my questions:

- Is the user-patterns functiontionality implemented in the tesserocr Python API of tesseract?

- How exact is the syntax of specifying user patterns with the tesserocr Python API. Is SetVariable() correct and how is the path (Linux) and the attribute specified?

- is there a default path, where it is lookes for the *.patterns / *.user-patterns file

With the attached code from my last message, I've tested different constellations with/without the combination of whitelist, different atrributes and path notations, which was not successfull. 

If I use the following notation for user patterns, it has no effect on the results independently from the entries of the *.patterns file:

 

api.SetVariable('user_patterns_file', '/home/roman/Dev_d/playground/user_patterns/deu.patterns')

Does anyone has (successfully) used user patterns with the tesserocr Python API of tesseract?

best wishes and thanks, Roman

[Zdenko Podobny]

Maybe I am wrong, but it looks to me like you are expecting from user-patterns something it never promises to provide.

What we know/experienced: 

• user-patterns extends the Tesseract legacy engine dictionary.
• putting a word/pattern to the Tesseract Legacy Engine dictionary never guarantees word is recognized correctly (see remark https://tesseract-ocr.github.io/tessdoc/APIExample-user_patterns.html)
• somebody (I can not find details as it was a long time ago) made tests and he found that the Tesseract legacy engine dictionary has limited effect. For "nonword" text (like "codes" with mixed letter&digits" people usually turn off the dictionary)
• some users prefer to use the Legacy engine for "codes" instead of LSTM

As far as I know, nobody made tests regarding LSTM and dictionaries e.g. an investigation if user-patterns also affect LSTM engine (as for LSTM there are new dictionary components lstm-punc-dawg, lstm-punc-dawg, lstm-number-dawg) ...

Zdenko

ne 3. 3. 2024 o 23:02 Roman Seidel <roman.s...@gmail.com> napísal(a):

--
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 view this discussion on the web visit https://groups.google.com/d/msgid/tesseract-ocr/CAL%3DSc5uQAOGF7dD%2BtP2xt93Phv9OYy6anDGLdar4gxZxEDwjYQ%40mail.gmail.com.

[Zdenko Podobny]

One correction:

I checked the example in the below mentioned url with the Tesseract executable and tessdata repository. The result is that user_pattern is effecting also LSTM. This could be easily tested by generating output without user_patters (Arial.txt):

tesseract Arial.png Arial

And with patterns:

tesseract Arial.png Arial.pat --user-patterns my.patterns

tesseract Arial.png Arial.pat.oem0 --user-patterns my.patterns --oem 0

tesseract Arial.png Arial.pat.oem1 --user-patterns my.patterns --oem 1

tesseract Arial.png Arial.pat.oem2 --user-patterns my.patterns --oem 2

Zdenko

ne 10. 3. 2024 o 17:32 Zdenko Podobny <zde...@gmail.com> napísal(a):