Community

XML output seems skewed despite straight image

Regarding the following task id: 37d7a2c1-0007-4b27-bace-0271c67c9fff

While the image is quite straight, the text baselines in the XML output indicate a significant rotation. For example, consider the following pieces of text and their baselines:

"1 Bloody Mary N/C" - 1143 "3.00" - 1176 "1 IBC Rootbeer" - 1265

From those results, you conclude that the "Bloody Mary" and "3.00" are on the same line more or less. However, looking at the image, it's clear that "3.00" is actually much closer to "IBC Rootbeer" than to "Bloody Mary" in terms of height.

Why might this be occurring, and how can we address it?

0

Comments

7 comments

  • Avatar
    G Moore

    I should add that we've actually been seeing this in a handful of other cases. Sometimes it affects most lines, as in the task indicated above, and sometimes it's localized and only seems to affect a couple of lines (e.g. top couple of item lines in task id 39501714-3649-478b-8adf-79cd38860ddd; compare this to a nearly identical image/task which didn't experience this issue, 58587abf-cfe7-4703-ac0a-06edf4936d77).

    My one guess is that this has something to do with different bend in the paper subject in the photos, but in all cases any bend is very minimal and really shouldn't have an effect

    0
  • Avatar
    G Moore

    Here's another one: 0bfd7aae-fe6f-4d22-9e20-e59dbb5c11ec Image is almost perfectly straight, but lines don't match up in the XML. Would really like this looked into, please.

    0
  • Avatar
    Anastasia Galimova

    Dear G Moore,

    The issue occurs because of the skew correction. As your images are skewed only a bit and coordinates are important for you, we recommend to turn it off.

    To turn off the skew correction, add "CorrectSkew=false" parameter to your request URL.

    0
  • Avatar
    G Moore

    I just tried this (task id 0c8a64c7-a787-4bc0-87d6-188005b874c4) and the result was no better. Now: "1 Bloody Mary N/C" - 971 "3.00" - 998 "1 IBC Rootbeer" - 1082

    As you can see the relative spacings between these lines is about the same as it was before, however like I said, the image clearly shows that the "3.00" line should be much closer to the "Rootbeer" line than to "Bloody Mary", which the XML output contradicts

    0
  • Avatar
    G Moore

    Anastasia, could you please follow up on this? Thank you

    0
  • Avatar
    Anastasia Galimova

    Dear G Moore,

    The developer says that to disable correction of all types of distortions it is necessary to set "imageSource=scanner" as well. In this case, coordinates are the following: "1 Bloody Mary N/C" - 974, "3.00" - 1040, "1 IBC Rootbeer" - 1083. Sorry for the inconvenience.

    0
  • Avatar
    G Moore

    Hmm I'll give it a try. However in this case, then, all of our input images will have to be perfectly level if no skew corrections at all will be made? Doesn't seem very robust.

    0

Please sign in to leave a comment.