Signature Extraction using Document Intelligence or Content Extraction

Adeniken Nathaniel 0 Reputation points

Can Azure Document intelligence or Content understanding extract signatures now? I know that this was not previously possible, but I'd like to know if it is possible to extract a signature from a document now? I know that this was not previously possible, but I'd like to know if it is possible to extract a signature from a document now

  1. Karnam Venkata Rajeswari 3,830 Reputation points • Microsoft External Staff • Moderator

    Hello @Adeniken Nathaniel ,

    Checking in to see if you had any chance to review the above response.

    Thank you

  2. Karnam Venkata Rajeswari 3,830 Reputation points • Microsoft External Staff • Moderator

    Hello @Adeniken Nathaniel ,

    Just checking in to see if you have got a chance to see my response to your question in resolving the issue.

    Thank you!

  3. Karnam Venkata Rajeswari 3,830 Reputation points • Microsoft External Staff • Moderator

    Hello @Adeniken Nathaniel ,

    Just checking in to see if you have got a chance to see my response to your question in resolving the issue.

    Looking forward to your response and appreciate your time on this.

    If the query has been resolved, please accept the answer by clicking the "Upvote" and "Accept Answer" on the post.

    Thank you!

Sign in to comment

2 answers

  1. Karnam Venkata Rajeswari 3,830 Reputation points • Microsoft External Staff • Moderator

    Hello @Adeniken Nathaniel

    Welcome to Microsoft Q&A .Thank you for reaching out to us.

    Signature extraction is a common requirement in document workflows, and it is important to clarify what is currently supported and how this can be implemented in a practical way using available Azure AI capabilities.

    At this time, prebuilt signature extraction is not supported in Azure AI Document Intelligence. None of the prebuilt models—such as Read, Layout, Invoice, Receipt, ID, or Contract - provide a dedicated “Signature” field or return the handwritten or drawn signature itself as a reusable image or object. These models are designed primarily for text, layout, tables, and structured field extraction rather than visual artifact extraction.

    Supported approaches are as follows-

    1. Custom Document Intelligence models for structured documents - Custom Template or Custom Neural models can be trained to detect and locate signatures.Please follow the following steps
      • Create a custom model in Document Intelligence Studio.
      • Upload representative sample documents containing signatures.
      • Label the signature area explicitly as a custom field (for example, “Signature”).
      • Train and deploy the model.
      At runtime, the model can:
      • Indicate whether a signature is present.
      • Return bounding box or polygon coordinates and confidence for the labeled signature field.
      Important limitation to be noted
      • The service does not return the signature image itself.
      • Any cropping or image extraction must be handled externally using the original document and the returned coordinates.
    2. Layout model for signature region identification for varied layouts - The Layout model extracts document structure such as lines, words, and bounding regions without applying business semantics. Practical usage:
      • Analyze documents using the Layout API.
      • Identify visual regions where signatures appear based on position and surrounding context.
      • Use returned coordinates to crop or process the region outside the service.
      This approach is suitable when document layouts vary significantly or when signatures are not tied to fixed templates. Limitations to be aware of are that
      • Signature verification, authenticity checks, or signature matching are not supported.
      • Signature extraction as a reusable image or vector is not returned by the service.
      • Any biometric or forensic analysis requires downstream computer vision or custom image processing solutions.

    Thus , for Signature extraction -

    1. Prebuilt signature extraction is not supported
    2. Signature presence and location detection is supported via custom models
    3. Region identification using layout analysis is supported
    4. Signature verification or matching is not supported

    The following references might be helpful , please check them out

    Thank you

    1. Manas Mohanty 17,185 Reputation points • Microsoft External Staff • Moderator

      Yes Adeniken Nathaniel

      Not supported on Prebuilt models yet as my colleague pointed out

      Please leverage Custom neural /extraction model to find location of signatures.

      Then you can compose the Custom neural/extraction model along classifier models to come with resilient solution that might fulfill the requirement.

      Note: Please follow the best practices mentioned in Transparency notes, Data and privacy and Code of conduct serving the solution to customers.

      Thank you.

    Sign in to comment

  2. AI answer

    Yes. Azure Document Intelligence now supports signature detection and extraction in multiple scenarios.

    Key options:

    1. Prebuilt bank check model
      The prebuilt US bank check model (v4.0, API version 2024-11-30, model ID prebuilt-check.us) supports signature detection on bank checks.
    2. Custom template models (v3.0+)
      Custom template (form) models support signature fields. Signature detection here means detecting the presence of a signature in the defined field, not identifying who signed. If the model returns unsigned, it did not find a signature in that field.
    3. Custom neural models (v4.0)
      The custom neural v4.0 (2024-11-30 GA) model explicitly supports signature detection, along with key‑value pairs, selection marks, tables, and overlapping fields.
      To use it:
      • Label fields with type Signature and draw the region around the signature.
      • Provide at least five training samples with labeled signatures (with some variation) for best accuracy.

    These capabilities allow extracting whether and where a document is signed (and the signature region), but they do not verify signer identity.

    Content Understanding is described as the evolution of Document Intelligence with broader multimodal capabilities, but the provided information does not add further signature-specific behavior beyond what is listed above.

    References:

    AI-generated content may be incorrect. Read our transparency notes for more information.
Sign in to answer

Your answer