The following are some additional tabular data tables in PNG format, with results returned from the Cloud Vision OCR engine below.
The first sample is a fairly clear inventory table with readable text.
The second sample is a less clear extract from an old book.
Overall, Google Cloud Vision seems like a very capable approach for quickly solving OCR tasks. Definitely worth keeping an eye on during project evaluation stages.
Next, I’m going to give Tesseract’s new LSTM support another try. Hopefully, they’ve made some progress towards a final release in the past few months.