How to recognize copyright symbol © using pre-built layout model

Question

How to recognize copyright symbol © using pre-built layout model

Rafael Gomes 0

When using the AI Document Intelligence service with the model prebuilt-layout, we are having issues recognising text containing the copyright symbol ©. Is there any possible solution(s) to fix/improve the accuracy for recognising this (and other) symbol(s)?

I've attached a simple document as an example of where this happens and what the Document Intelligence Studio outputs when analysing it.

test-symbol.pdf

Document Intelligence Studio analysis result when trying to recognize copyright symbol ©

santoshkc 15,590 Reputation points Microsoft External Staff Moderator

2025-07-01T13:46:28.6766667+00:00

Hi Rafael Gomes,

Following up to see if the given response was helpful. Incase if you have any queries do let us know. Thank you.
Rafael Gomes 0 Reputation points

2025-07-01T14:40:39.5933333+00:00

Thank you for your quick reply @santoshkc . Given that we are already using the recommended model and using high-resolution feature, I guess there is not much more we can do for now to improve the model's accuracy in symbol(s) detection.

Do you know if it is on the short-term roadmap to improve the prebuild-layout model on this topic? What is the expected release cycle for new model versions? Is it worth us sending example documents on where this symbol is not correctly detected to the reported email address (this was presented on the Azure portal support)?
Pavankumar Purilla 10,350 Reputation points Microsoft External Staff Moderator

2025-07-02T17:00:00.9566667+00:00

Hi Rafael Gomes,
Since you are already using the recommended model along with the high-resolution feature, there may be limited additional configuration options currently available to improve symbol detection accuracy within the Azure AI Document Intelligence prebuilt-layout model. At this stage, further improvements would likely depend on enhancements made by Microsoft to the underlying model itself. While Microsoft does not publicly disclose a detailed short-term roadmap for specific improvements such as symbol recognition, their models are regularly updated, typically with minor updates every few months and major updates on an annual or semi-annual basis. These updates are aimed at improving overall OCR, layout understanding, and content extraction accuracy. Given this, it would indeed be worthwhile to share example documents where the model fails to detect specific symbols by sending them to the provided contact email address, which is listed in the Azure portal support documentation. Submitting these examples helps Microsoft identify and analyze edge cases, contributing to future model improvements and potentially accelerating the development of features that address your specific needs.

1 answer

Your answer

santoshkc 15,590 Reputation points Microsoft External Staff Moderator

2025-07-01T13:46:28.6766667+00:00

Hi Rafael Gomes,

Following up to see if the given response was helpful. Incase if you have any queries do let us know. Thank you.
Rafael Gomes 0 Reputation points

2025-07-01T14:40:39.5933333+00:00

Thank you for your quick reply @santoshkc . Given that we are already using the recommended model and using high-resolution feature, I guess there is not much more we can do for now to improve the model's accuracy in symbol(s) detection.

Do you know if it is on the short-term roadmap to improve the prebuild-layout model on this topic? What is the expected release cycle for new model versions? Is it worth us sending example documents on where this symbol is not correctly detected to the reported email address (this was presented on the Azure portal support)?
Pavankumar Purilla 10,350 Reputation points Microsoft External Staff Moderator

2025-07-02T17:00:00.9566667+00:00

Hi Rafael Gomes,
Since you are already using the recommended model along with the high-resolution feature, there may be limited additional configuration options currently available to improve symbol detection accuracy within the Azure AI Document Intelligence prebuilt-layout model. At this stage, further improvements would likely depend on enhancements made by Microsoft to the underlying model itself. While Microsoft does not publicly disclose a detailed short-term roadmap for specific improvements such as symbol recognition, their models are regularly updated, typically with minor updates every few months and major updates on an annual or semi-annual basis. These updates are aimed at improving overall OCR, layout understanding, and content extraction accuracy. Given this, it would indeed be worthwhile to share example documents where the model fails to detect specific symbols by sending them to the provided contact email address, which is listed in the Azure portal support documentation. Submitting these examples helps Microsoft identify and analyze edge cases, contributing to future model improvements and potentially accelerating the development of features that address your specific needs.

Answer 1

Hi Rafael Gomes,

Thank you for reporting this. We were able to reproduce the issue using the prebuilt-layout model (2024-02-29 Preview) and observed that the copyright symbol © is being recognized as the letter “C”. This behaviour is expected in some cases, as the model primarily focuses on extracting general text and may misinterpret certain special characters based on document quality, font style, or encoding.

As a workaround, we recommend continuing to use the latest prebuilt-layout model version (2024-02-29 Preview), which has improved accuracy compared to previous versions for special characters, though it may still have occasional limitations. Additionally, ensuring the document has high resolution, clear fonts, and good contrast can help improve recognition.

User's image

I hope you understand! Thank you.

Share via

How to recognize copyright symbol © using pre-built layout model

1 answer

Your answer