It has a hard time with the letter Q for some reason now. Thanks for your
help, it seems just using the included traineddata is more accurate than
just the font, don't know why that is.
Thanks for letting me know. No I haven't had a chance. I will try 4.0
although I have never manually dealt with tesseract. I've been using
programs for 3.x that trained and made box files automatically.
On Apr 24, 2017 12:43 AM, "ShreeDevi Kumar" wrote:
> James,
>
> Were
James,
Were you able to get this to work for you with 3.04/3.05?
I get accurate results using Tesseract 4.0 alpha, though it takes longer
with --oem 1 than --oem 0.
./troublewith98-300.jpg
Tesseract Open Source OCR Engine v4.00.00alpha-385-gab41465 with Leptonica
real0m1.203s
user
Try training using more samples of 8, 9, B etc.
What results do you get with the provided eng.traineddata? Are they better
or worse?
Have you tried changing DPI of image to 300?
- excuse the brevity, sent from mobile
On 22-Apr-2017 10:29 PM, "James Abney" wrote:
> Oh yes
Oh yes I guess I forgot to include that information, I did train using only
that font and with the same size font. I am on windows 7 and I used 3.05 to
train, although the .net wrapper i use is 3.04. I don't see how it has
difficulty with the 9 and 8, seems very odd.
On Friday, April 21, 2017
Which version of Tesseract. Which o/s?
If all your text is in tungsten-semibold, have you tried training with just
that font?
- excuse the brevity, sent from mobile
On 22-Apr-2017 12:50 AM, "James Abney" wrote:
The font is tungsten semibold
On Friday, April 21, 2017 at
6 matches
Mail list logo