Azure Form Recognizer is now Azure AI Document Intelligence with new and updated capabilities
Published Jul 18 2023 09:00 AM 31.1K Views
Microsoft

Azure AI Document IntelligenceAzure AI Document Intelligence

In today's digital era, businesses generate and consume an immense amount of information contained within documents. Unlocking the insights and knowledge hidden within these documents is crucial for organizations striving to enhance productivity, automate processes, and gain valuable insights. That's why we are thrilled to share our revamped product, Azure AI Document Intelligence, formerly known as Azure Form Recognizer. With its enhanced capabilities and expanded scope, Document Intelligence is set to empower organizations across all industries to harness the full potential of document information.

Document Intelligence is designed to help organizations streamline their operations and boost productivity. it enables efficient extraction of data from diverse document formats. From invoices and receipts to contracts and reports, Document Intelligence is capable of swiftly and accurately capturing information, eliminating the need for manual data entry. This time-saving automation allows employees to focus on more strategic tasks, resulting in increased productivity across the board.

 

Latest updates

The service continues to evolve and improve based on your feedback. This month we will be adding new and updated capabilities to Document Intelligence.

 

Document classification and splitting

A common challenge for customers processing documents on a scale is processing multi document sets. With the addition of document classifiers, we now support most common scenarios with multiple documents.

Document Intelligence Studio - custom classificationDocument Intelligence Studio - custom classification

  1. Document identification: When you have customers uploading documents which needed to be routed to the appropriate model for extraction, the document classification model can now identify the document to support conditional routing workflows.
  2. Multi document portfolios: Common in scenarios like mortgage applications where a single file consists of multiple individual documents. With the new document classification and splitting API, you can identify the page ranges for the individual documents and route them to the appropriate extraction model.
  3. Multiple instances of the same document: In scenarios like invoice processing, its common to receive a single file containing multiple invoices. The classification and splitting API can classify and split multiple instances of the same document type for downstream processing.

Training a classifier in the Studio is extremely simple and if you have your samples organized in folders, you could train a classifier with just a few clicks. You can now configure more complex document processing workflows that rely on classification splitting and extraction.

 

New prebuilt models for contracts and tax forms

Prebuilt models provide a high-quality response with a defined schema for known document types. Contracts are complex legal documents common to document processing scenarios like contract lifecycle management, procurement and legal services.

Document Intelligence Studio - prebuilt contractDocument Intelligence Studio - prebuilt contract

With the new contract prebuilt model, you can now extract the common fields from any contract with a simple API call.

The release also includes prebuilt models for the 1098 tax form variants, including the 1098, 1098-E and 1098-T. With the existing W-2 prebuilt model the 1098 models support further automation of tax document processing for tax and financial services workflows. Document Intelligence Studio - prebuilt 1098Document Intelligence Studio - prebuilt 1098

 

 

 

 

Content and structure extraction

The July GA release will include updates to content extraction with new add on capabilities.

  1. Support for extracting barcodes from documents. Healthcare and procurement have many document types where critical information like patient id is encoded in the barcode. Adding support for barcodes to all models, you can now build the automation solution to extract and process barcodes from documents.Document Intelligence Studio - layout barcode extractionDocument Intelligence Studio - layout barcode extraction
  2. High resolution documents are common in industries like heavy engineering. With added support for high resolution documents, you can now extract content from these high-resolution documents.
  3. Formulas can now be extracted from documents in LaTeX format. Formulas a common component of documents in scientific research, with this new addition to the content extraction capability, you can now recognizer and extract formulas in your workflows. Document Intelligence Studio - layout formula extractionDocument Intelligence Studio - layout formula extraction
  4. Fonts is the latest addition to the structure extraction capabilities that include paragraphs, paragraph roles, selection marks and tables to name a few. Fonts can be useful when creating a high-fidelity digital version of a scanned document or a feature in downstream extraction and segmentation tasks.
  5. AI quality updates for table extraction, improvements to single character text recognition and handwritten text recognition improvements are among the many improvements in all the models.

 

 

Custom model updates

 Build a custom model to extract a specific schema from any document or form. This release brings a few enhancements to custom models.

  1. Custom neural models will now support 47 languages including Chinese, Japanese, Korean and Arabic. You can train a custom neural model for documents in any of the supported languages to extract the fields needed.
  2. Training a custom neural model just got a lot easier. Train a neural model with just one labeled document! Custom neural models can now be trained with a single labeled document.
  3. Custom template models are also updated with improvements to the signature detection capability.

 

Studio

  1. New document classification model train and test experience. Training a classification model in the Studio is as simple as selecting the documents, assigning the labels and training the model.
  2. Developer productivity improvements with human in the loop (HITL) for custom model training. With the human in the loop workflow, the label, train and test loop is now tighter and more effective. Identifying the documents that specifically result in lower accuracy for labeling improves the model performance.Document Intelligence Studio - custom auto labelDocument Intelligence Studio - custom auto label
  3. Auto labeling documents with a prebuilt or custom model further reduces the effort involved in adding a document to the training dataset. Pre-labeling ensures that you are only editing the few field(s) where the model predictions did not meet expectations.
  4. Improvements to dataset management include finding and filtering documents in the training dataset.

 

Get started with Document Intelligence today!

With the renaming of our product to Azure AI Document Intelligence, we are expanding its scope and reinforcing its mission: to empower organizations across industries to unlock the power of information. By enabling increased productivity, automating business processes, and extracting knowledge and insights, Document Intelligence serves as a catalyst for digital transformation. Join us on this exciting journey as we empower your organization to leverage the hidden potential within documents and thrive in the data-driven world.

Here are a few links to learn more about Document Intelligence:

  1. Start with the documentation to learn more about the different capabilities.
  2. Test the different models with sample documents or your own documents in the Studio

The Studio is your single stop to experience the all the capabilities within the service. Try using any of the general document or prebuilt models to extract content, structure and fields or train and test a custom model for classification or extraction. Aligned with this release, the Studio adds new capabilities specifically intended to improve developer productivity with training custom models.

 

1 Comment
Co-Authors
Version history
Last update:
‎Jan 25 2024 08:02 AM
Updated by: