OCR numbers

[hsl20130659]

it seems that the model can not recognize numbers

[ChWick]

Which model did you use?

[hsl20130659]

fraktur_19th_century

[ChWick]

This model is trained to predict both characters and numbers. Could you please provide a line of your data where it fails?

[hsl20130659]

it writes nothing in the result.txt

[hsl20130659]

I use a picture that has numbers and chinese characters

[ChWick]

The fraktur 19th century model can only be used for Fraktur fonts: The characters must be similar to https://en.wikipedia.org/wiki/Fraktur. It might be possible that the model could possibly detect Arabic digits embedded in Chinese fonts, however I can not guarantee that. Do you input a full image or only a single horizontal text line? The latter is required.

[hsl20130659]

I put a full image into the network ,my goal is just recognizing the numbers and I dont care if there are chinese characters or not.

[ChWick]

Calamari can only process single lines, not a full image. You need a separate tool to handle the line segmentation. You can use e. g. https://github.com/tmbdev/ocropy/wiki/Page-Segmentation for the page and line segmentation.

[hsl20130659]

my pictures are cropped from ID cards ,the row of birthdays

[ChWick]

Solely the row of birthdays? Can you attach a single image file. Maybe you need https://github.com/Calamari-OCR/calamari_models/tree/master/antiqua_modern as model (Modern typographic fonts).

[hsl20130659]

``

[hsl20130659]

you can test the pictures,and show me the results

[ChWick]

This might work, but additional steps are required:

• Crop the bottom area to the content with approximately 3px margin around the text.
• Binarize the image

I manually did those steps and got (. 1986t 1 t 8 fH as result for 001.png.

[ChWick]

My manually processed input line:

[hsl20130659]

do you mind I make a friend with you ? My qq is 913452977, I think I have something more to ask you ,hehe

[ChWick]

You can E-Mail me via christoph.wick@informatik.uni-wuerzburg.de.