I was going to use a free account api to scan crosswords like https://i.stack.imgur.com/qqy4X.png but I don’t get anything sensible out from the front page tool.
Is there anything else I could do?
This is an interesting test case. The challenge here is the number OCR, not the table capture itself. The ocr engines miss some numbers.
As of right now, you could combine the results of engine1 and 2, this way you will get more numbers, but unfortunately still not all.
The good news is that we are working on further improved number OCR and will have an updater later this year.
Preview: Our internal ocr beta version already catches almost all numbers, except the single digit “7” in the top row:
PS: The numbers in the OCR screenshot mark the sequence of OCR’ed words (here: numbers). They are not the detected values itself.
1 Like
Ahh, so is the grid captured or not? My full intention is to use https://github.com/dwmkerr/crosswords-js to turn an image into a more accessible crossword, so I’d need to be able to figure out the x,y positions eventually
The x/y positions of every number (word) are returned, already now.
OCR captures words, not the grid itself. But since you have the x/y of each number, you also have the position of the grid.