READ OCR Bounding Box Accuracy

Baruch Shadrouz 31 Reputation points
2021-03-08T15:59:38.083+00:00

I am using the basic Code for the READ OCR that Microsoft gives in their getting started guide. The only modification is that I am running that program with multiprocessing with 4 cores so I am making multiple calls to the API at the same time.

The READ OCR sometimes gives me a Bounding Box of a word that is either too large or too small for that word. You can see in the screenshot below that the word "40.00" was completely found but the bounding box is too small and does not cover the entire word. The same issue happens with the word "Current". There is cases where the bounding box is too large as well.
75542-image.png

Any help would be appreciated!

Azure Computer Vision
Azure Computer Vision
An Azure artificial intelligence service that analyzes content in images and video.
338 questions
Azure AI services
Azure AI services
A group of Azure services, SDKs, and APIs designed to make apps more intelligent, engaging, and discoverable.
2,577 questions

5 answers

Sort by: Most helpful
Most helpful Newest Oldest
  1. Mitchell 11 Reputation points
    2022-08-01T11:26:15.517+00:00

    I have found the same as well, the below image shows Tesseract bounding boxes in Blue and Azure Read API bounding boxes in Red. Almost every bounding box from azure is taller than the actual text and nearly all shifted to the left slightly.

    Is there any update for a fix to this issue? @GiftA-MSFT

    226665-azure-vs-tesseract.png

    2 people found this answer helpful.
    0 comments
  2. cr 6 Reputation points
    2022-12-03T08:20:36.987+00:00

    I am struggling with the same problem as @Mitchell . Bounding boxes are inaccurate and often shifted to the left.
    Is there any progress on this? Can we expect this problem to be fixed at some point? @GiftA-MSFT
    As my use case is depending on exact bounding boxes, I'd need to find another solution if fixing this is not on the roadmap.

    Thanks!

    1 person found this answer helpful.
    0 comments
  3. dj-ancora 1 Reputation point
    2023-01-03T00:43:16.48+00:00

    I too am have the same problem with the bounding boxes often being too spacious on the left/top/bottom sides and not properly encompassing the text on the right side.

    0 comments
  4. Anurag Singh 0 Reputation points
    2024-04-26T07:23:28.19+00:00

    I also have same problem. in most of documents 1, D and 5 is extracting wrong.

    0 comments
  5. Nico 15 Reputation points
    2024-06-03T11:07:00.7233333+00:00

    Dear all,

    I see the same problem, both for the Azure Read API V3 and the Azure Document Intelligence Read API.

    @Microsoft-Team, can you please give an update on this (do you consider this as a problem, will or won't you fix this)?

    Thanks and all the best,
    Nico

    0 comments