More incorrect Cloud OCR results lately?

  • Last Post 14 February 2017
G Moore posted this 13 February 2017

I've been noticing more incorrect digits in our Cloud OCR results lately. They're cases which are pretty clear-cut and should be easily recognized. Lots of 3's instead of 9's and some 5's instead of 6's.

Here's one example:


The 6 is recognized as a 5. The char rec variants don't even include 6 as one option. Seems very wrong.

Another example:


Both of the trailing 9's are interpreted as 3's.

In the past, I feel like the API would have handled all of these cases just fine. Has something changed?

Order By: Standard | Newest | Votes
Oksana Serdyuk posted this 14 February 2017

We haven't changed anything that may affect your results. Moreover, I have processed these two images using the processImage method with the profile=*textExtraction and imageSource=*photo parameters and the results are fine:

alt text

So, would you be so kind to send the source images to for our testing on our side?

Attached Files

G Moore posted this 14 February 2017

Sure, have just emailed these two images! Thanks!