OCR on Burnins?

Would love to extract shot names from a stringout… Does anyone know of a tool that does OCR on burnins?

I don’t know of an existing tool, but since you added a python tag, I assume maybe you are interested in rolling your own?

I’ve done some light work with tesseract in python, which does a pretty good job of recognizing text in images. So, I think all the pieces are there to build something. How much it can be automated depends on a lot of things, but I would imagine the tasks to be kind of like this:

  • Flame scene detect to cut a single clip into multiple clips/timeline segments wherever the burnin changes (assuming scene detect would be sensitive enough for this)
  • Get a still frame from each segment (Flame’s python hooks can do this, I think?)
  • Use tesseract in python to analyze the image for text and return the results
  • Do something with the results, like compiling them into a text file with timecode or building textFX on top of each segment with the OCR text, etc.

I’m betting some of the python pros here might be able to hack something together to do this or perhaps make something that gets close. I don’t have much experience with python and less with python in Flame but it seems like all or some of this should be within reach.

-Matt

Matt! I think your outline is dead on. I’d add the ability to designate a ROI to speed up tesseract’s accuracy or at least single out which burnin yields data.

Finn had some good suggestions on Discord, but all signs point to tesseract.