Test cases

[falktan]

A small collection of sample pictures to test the quality of the OCR algorithm would be helpful.
I.e. there should be about 15 pictures with different lighting conditions and different text size and quality with more or less distracting background.
The repo https://github.com/falktan/ovip-supplementary is probably a good place for these pictures.

In addition, it would be helpful to have a convenient way to test the algorithm on those pictures.

[falktan]

It might also be a good idea to search for a reasonable collection of sample images that is available publicly.

[falktan]

This paper references a number of sources for "out in the wild" text (as a bonus it is an interesting read about how to use LSTM for OCR).
https://arxiv.org/pdf/1507.05717.pdf

[falktan]

Started work here:
https://github.com/falktan/ocrjs

[falktan]

Ideas how to improve tesseract see:
https://tesseract-ocr.github.io/tessdoc/ImproveQuality.html