Stroke Width Transform

The latest version of OpenOCR has the ability to preprocess the image using Stroke Width Transform, which is capable of extracting the text from an image.

Here is a visual example of Stroke Width Transform in action.

Running it

This was just recently added, and so the launcher.sh script hasn't been updated yet. In the meantime, you should be able to get it running with the following:

$ export AMQP_URI=amqp://admin:${RABBITMQ_PASS}@${DOCKER_HOST}/
$ docker run -d tleyden5iwx/open-ocr-preprocessor open-ocr-preprocessor -amqp_uri "${AMQP_URI}" -preprocessor "stroke-width-transform"

Verifying it works

$ curl -X POST -H "Content-Type: application/json" -d '{"img_url":"http://bit.ly/ocrimage-swt","engine":"tesseract", "preprocessors":["stroke-width-transform"]}' http://${DOCKER_HOST}:${HTTP_PORT}/ocr

Expected result:

YH XMCDMTDC

Compare to default Tesseract output

$ curl -X POST -H "Content-Type: application/json" -d '{"img_url":"http://bit.ly/ocrimage-swt","engine":"tesseract"}' http://${DOCKER_HOST}:${HTTP_PORT}/ocr

Expected result:

E' ,‘YHwacpMTDCH ;
 3?". ‘  V‘L"~m> I shah-r}. I’VMU' i 5: 1“”. A"

As you can see, in this particular case the Stroke Width Transform makes a huge positive difference.

Black-on-white text

By default, it expects black text on a white background. However, if you have white text on a black background, you will want to pass an additional parameter as follows:

curl -X POST -H "Content-Type: application/json" -d '{"img_url":"http://bit.ly/ocrimage-swt","engine":"tesseract", "preprocessors":["stroke-width-transform"], "preprocessor-args":["0"]}' http://${DOCKER_HOST}:${HTTP_PORT}/ocr

Legal values:

• "0" -- white text on a black background
• "1" -- black text on a white background (default)

In the case of this test image, since it's black test on a white background, passing in "0" completely breaks the OCR and it returns no output.

References

• Stroke Width Transform - alternate link
• Project Naptha