Roulette Ideas

This thread is for the discussion of RIBOT's OCR ways for numerical recognition.

There are many techniques possible in order to achieve the goal of capturing numbers from saved images and live online casinos into actual betting modules.

From traditional 1-to-1 pixel comparison to free libraries like Tesseract OCR and we can also evaluate online APIs.

Each approach has pros and cons.

We are going to explore them in the open since input modules are entirely free for public usage on RIBOT FREE version.

Feel free to attach sample images with roulette numbers (from casinos, Excel sheets, other) in this thread.

We are scheduled to complete the image-to-text-file OCR module to close the month 👍

This code is being done as a RIBOT module & also as a web app for our server 👍

Update: An "OCR definition" format is being developed to handle numerical rectangles, black&white mode, color saturation/grayscale + image reading orientation.

Update: we are deferring the web app to focus on a Windows OCR with Tesseract v4.1 and the good ol' image slicing technique to feed the "number squares" by coordinates.

Users will have to download Windows binaries from:

https://digi.bib.uni-mannheim.de/tesseract/

Yet another update: the web app is being retaken with Tesseract JS (which is a JavaScript port of the OCR engine that runs in the browser).

The page for it is:

https://rouletteideas.com/ocr/

OCR definitions are public; starting with the Hippodrome definition as per the above.

UPDATE:

- Making tests with the GOCR engine.
https://antimatter15.com/gocr.js/demo.html

- Testing number rectangles via image clipping (slow) and via Tesseract's "SetRectangle".
https://rdrr.io/github/duncantl/Rtesseract/man/SetRectangle.html

The approach that "wins" becomes definitive and we call Hippodrome OCR complete.

Quote from: VLS on Sep 19, 2023, 08:19 PMThe approach that "wins" becomes definitive and we call Hippodrome OCR complete.

Rectangle clipping + Tesseract won!

Numbers are being recognized correctly now.

The canonical images are the ones by Hardman, as per the above:

https://rouletteideas.com/index.php?action=dlattach;attach=299

...It is a matter of setting all the number rectangles' coordinates (perhaps with a bit of extra color processing).

Hippodrome is +90% done 👍

---

By the way, the /ocr/ pretty URL still needs to be enabled on the new server (so many details!). I forgot to mention the OCR app URL is currently available via the full link:

https://rouletteideas.com/ocr/index.html

I'll add the pretty URLs rules in the morning, it is past midnight here.

---

Thanks for your patience once again. We are delivering ALL due programs now, one by one.

Good day Guys!

Just re-enabled pretty URL @ https://rouletteideas.com/ocr/

Time to set all the tiny rectangles...

The OCR page is actively being worked on.

This is being addressed on the weekend:

- Bottom-right to top-left numerical order.

- Manually testing different color operations toward achieving pure 2-bit "black & white" transformation for most accurate recognition.

- Generating a downloadable text file, having one number per line as it's customary for roulette actuals files.