Pretty bad result for non dictionary words

[Christian Koch]

I try to recoginze product codes written in images.

The results in tesseract 3.04.00 are pretty bad. Even when I try a primitive example (see attachment) it won't work.

Instead "ABC-DEF" I get "AECVDEF"

The example works flawlessy in gocr but guess I'm just using wrong settings or something similar in tesseract.

Or is tesseract in general not designed to recognize "random strings" and I should rather use another tool (recommendations?) for this case?

[Rolf Mertig]

If you resize with convert from ImageMagick (or any other tool):

convert ocr.jpg -resize 150% ocr2.jpg

then 

tesseract ocr2.jpg ocr2 ; cat ocr2.txt

gives 

ABC-DEF

[Christian Koch]

Hi Rolf,

thank you for your response.

Is this the "right" way? I read that I should rather use proper settings in tesseract than doing manual processing.

Are smaller texts a problem in general?

[Tom Morris]

On Thursday, May 12, 2016 at 5:39:23 PM UTC-4, Christian Koch wrote:

Are smaller texts a problem in general?

Yes. https://github.com/tesseract-ocr/tesseract/wiki/FAQ#is-there-a-minimum-text-size-it-wont-read-screen-text 

Tom