How to recognize random transformed digits from images with tesseract + ML.

Requirements

tesseract >= 4.0

tesstrain : https://github.com/tesseract-ocr/tesstrain

python

Training image set

images : ground_truth/.png
correct_text : ground_truth/.txt

Image files should be png type.

checking image set

Do not reuse images from the training set.

It needs to use different images from the training set.

Sample ground truth image

Sample image : 36-0001.png

Sample image : 36-0002.png

Sample image : 36-0003.png

Files prep

bestdata folder and python files are that getting from the tesstrain project.

.
├── bestdata
│   ├── eng.traineddata
│   ├── ....
│   ├── script
│   │   ├── Arabic.traineddata
│   │   └── ....
│   └── ...
├── data
│   └── my-model-ground-truth
│       ├── 0-0001.gt.txt
│       ├── 0-0001.png
│       ├── 0-0002.gt.txt
│       ├── 0-0002.png
│       ├── 0-0003.gt.txt
│       ├── 0-0003.png
│       └── ....
├── generate_gt_from_box.py
├── generate_line_box.py
├── generate_line_syllable_box.py
├── generate_wordstr_box.py
├── normalize.py
├── requirements.txt
├── shuffle.py
└── tree.txt

Training the model

make training MODEL_NAME=my-model START_MODEL=eng TESSDATA=./bestdata

How long takes time to make model?

system : Intel(R) Core(TM) i5-7300HQ CPU @ 2.50GHz

image size : 160x80

image nums : 1000 images

recognize target : signle digit

real	9m54,756s
user	29m45,056s
sys	0m31,379s

Output

./data/my-model.traineddata

copy your traineddata to tesseract tessdata

cp ./data/my-model.traineddata /usr/share/tesseract-ocr/4.00/tessdata/

Test

tesseract 36-1003.png --psm 8 -l my-model

References

https://stackoverflow.com/questions/43352918/how-do-i-train-tesseract-4-with-image-data-instead-of-a-font-file

https://github.com/tesseract-ocr/tesstrain

https://github.com/tesseract-ocr/tessdata_best

tesseract help

Page segmentation modes:
  0    Orientation and script detection (OSD) only.
  1    Automatic page segmentation with OSD.
  2    Automatic page segmentation, but no OSD, or OCR.
  3    Fully automatic page segmentation, but no OSD. (Default)
  4    Assume a single column of text of variable sizes.
  5    Assume a single uniform block of vertically aligned text.
  6    Assume a single uniform block of text.
  7    Treat the image as a single text line.
  8    Treat the image as a single word.
  9    Treat the image as a single word in a circle.
 10    Treat the image as a single character.
 11    Sparse text. Find as much text as possible in no particular order.
 12    Sparse text with OSD.
 13    Raw line. Treat the image as a single text line,
       bypassing hacks that are Tesseract-specific.

OCR Engine modes: (see https://github.com/tesseract-ocr/tesseract/wiki#linux)
  0    Legacy engine only.
  1    Neural nets LSTM engine only.
  2    Legacy + LSTM engines.
  3    Default, based on what is available.