how to use tessdata_best for tesseract (pytesseract). What are the arguments and procedure?

Question

TL;DR: How do I install tessdata_best to use withpytesseract inside conda in Ubuntu 18?

I have been using pytesseract inside conda environment for quite some but there is a need to improve the accuracy and I found out that tessdata_best gives you the best accuracy. How can I install and use that version? I am using Ubuntu 18 and have to work with pytesseract.

I have my tesseract installed at /usr/share/tesseract-ocr/ and inside it there is only 1 tessdata.

Do I need to get the tessdata_best from github by copying it to the directory /usr/share/tesseract-ocr/ alongside tessdata?

Even then, if I want to use tessdata-best, what do I have to use? Do I need to change the config as --oem 0/1/2/3?

Third and last thing is that I have my language.trainedata files at /home/deshwal/anaconda3/envs/py36/share/tessdata/eng.traineddata. Do I need to paste the tessdata_best at this location too? Becuse when I try to change the language dir, it gives me error as as:

/home/deshwal/anaconda3/envs/py36/share/tessdata/equ.traineddata Please make sure the TESSDATA_PREFIX environment variable is set to your "tessdata" directory. Failed loading language \'equ\' Tesseract couldn\'t load any languages! Could not initialize tesseract.'

Contraindicate

If you have found an answer, please let me know. If you have not, try to look into re-training Tesseract on new images -- it's quite a learning curve, but I think it will be worth it in the end. Best of luck! – Dutch

Have you found an answer? I want to use the tessdata_best but I don't know how. – Gluttony

@Gluttony No luck at all till now – Contraindicate

Answer 1

According to the documentation of pytesseract, you can use config argument with --tessdata-dir, as follows :

# Example config: r'--tessdata-dir "C:\Program Files (x86)\Tesseract-OCR\tessdata"'
# It's important to add double quotes around the dir path.
tessdata_dir_config = r'--tessdata-dir "<replace_with_your_tessdata_dir_path>"'
pytesseract.image_to_string(image, lang='chi_sim', config=tessdata_dir_config)

For more details see https://pypi.org/project/pytesseract/.

Bracket

Answer 2

I dont know if I understand your question clearly, however let me know if below helps ... You need to set datapath with location where you will copy the tessdata_best training models, For example,

Tesseract tesseract = new Tesseract(); // JNA Interface Mapping tesseract.setDatapath("/home/tesseract/tessdata_best_4_0_0/tessdata");

All your .traineddata files which you downloaded from (https://github.com/tesseract-ocr/tessdata_best) should be placed in the directory you define in setDataPath (for example:, /home/tesseract/tessdata_best_4_0_0/tessdata).

Please note: These models only work with the LSTM OCR engine of Tesseract 4 so make sure you have used library 4.1 or above.

Regards, Maulik

Adrien