user-words / bazaar

[Stef]

I'm trying to use user wordlists with the bazaar config but it seems to have no effect on the OCR result in my case. Therefore I printed the current parameters to verify whether the user-words list is used. This confirmed that the variables
load_system_dawg, load_freq_dawg, user_words_suffix and user_patterns_suffix were set correctly (to 0, 0, user-words, and user-patterns, respectively) but it doesn't show the user-provided filenames:

user_words_file        A filename of user-provided words.
user_patterns_file        A filename of user-provided patterns.

Is this just a bug/omission in the --print-parameters output or did tesseract fail to load these files (tessereact 3.05.00dev)?

[Meh Hem]

Hi Stef,

They have indeed no effect as far as I have found. The idea is great, but unfortunately it just does not seem to work. 

I have found no working demonstrations of it after looking for quite an amount of time.

We have instead found a strong ambiguous character set combined with processing the output helps to get results within an expected pattern.

[Tom Morris]

Works for me.  What command line are you using?

$ tesseract --user-words foo --user-patterns bar --print-parameters | grep user_

Error: failed to load foo

Error opening pattern file bar

Error: failed to load bar

user_words_file foo A filename of user-provided words.

user_words_suffix A suffix of user-provided words located in tessdata.

user_patterns_file bar A filename of user-provided patterns.

user_patterns_suffix A suffix of user-provided patterns located in tessdata.

[Stef]

Tom,

I wasn't aware of the new possiblity to specify user words on the command line. Instead I used the config file method with the following command lines and outputs:

tesseract.exe --version
tesseract 3.05.00dev
 leptonica-1.72
  libgif 4.1.6(?) : libjpeg 8d (libjpeg-turbo 1.3.1) : libpng 1.6.17 : libtiff 4.0.3 : zlib 1.2.8 : libwebp 0.4.3

tesseract.exe test.jpg stdout -l deu --print-parameters bazaar   | grep load_system\|load_freq\|user_
load_system_dawg    0    Load system word dawg.
load_freq_dawg    0    Load frequent word dawg.

user_words_file        A filename of user-provided words.

user_words_suffix    user-words    A suffix of user-provided words located in tessdata.
user_patterns_file        A filename of user-provided patterns.
user_patterns_suffix    user-patterns    A suffix of user-provided patterns located in tessdata.


tesseract.exe test.jpg stdout -l deu --print-parameters   | grep load_system\|load_freq\|user_
load_system_dawg    1    Load system word dawg.
load_freq_dawg    1    Load frequent word dawg.

user_words_file        A filename of user-provided words.

user_words_suffix        A suffix of user-provided words located in tessdata.

user_patterns_file        A filename of user-provided patterns.

user_patterns_suffix        A suffix of user-provided patterns located in tessdata.

My bazaar config file:

load_system_dawg     F
load_freq_dawg       F
user_words_suffix    user-words
user_patterns_suffix user-patterns

For the time being, I solved my problem by increasing the scan resolution from 300 dpi to 600 dpi which ensures everything to be recognized correctly with the default (system) settings.

[Tom Morris]

Perhaps this is just a misunderstanding or bad documentation.  The --print-parameters dump shows the input parameters, and the user_words_file / user_patterns_file parameters, if they're not set on the command line, will always be empty.

The actual file name that gets loaded gets computed on the fly here:

    https://github.com/tesseract-ocr/tesseract/blob/master/dict/dict.cpp#L274
but the result isn't saved into the user_words_file parameter

Tom

[Stef]

Thanks for this clarification.

Stef