Have you tried Google Cloud Vision at all - its OCR seems superior to
Tesseract from tests I have done to date.
I just span up a project for you and it did pretty well (error on single
digit zero which matched \n2 instead of 0) but maybe with some of the
preprocessing you are doing it will work be
Tensorflow? No, no, no. You've switched from a chainsaw to a sledgehammer,
your flowers are going to be exceedingly unhappy! I just had a Googler
present Tensorflow to the GDG I run a couple of weeks ago and if we were
working with handwritten digits, we could get you up to 99+% accuracy in a
m
Hey Tom,
These are video screenshots, so no luck on the EXIF info, that was the
guiding idea. Your feedback was exactly what I was looking for. Sounds like
tesseract isn't the right tool, on to tensorflow and building my own deep
learning structure.
Thanks!
Ben
On Saturday, December 31, 201
p.s. Taking an even bigger step back - are you sure there isn't digital
metadata available either in a separate file or embedded in the EXIF data
for the image? The whole image processing schtick, while fun, may be
unnecessary.
Tom
On Saturday, December 31, 2016 at 12:29:02 PM UTC-5, Tom Morri
I'm not sure Tesseract is the tool for the job here. It strikes me as the
CV equivalent of taking a chainsaw to prune your flower garden.
Are the captures all from the same model camera? A few different models?
The one example image that you showed uses a low resolution, probably fixed
pitch, f
5 matches
Mail list logo
