Document Intelligence public preview releases provide early access to features that are in active development. Features, approaches, and processes may change, prior to General Availability (GA), based on user feedback.
The public preview version of Document Intelligence client libraries default to REST API version 2024-07-31-preview.
Public preview version 2024-07-31-preview is currently only available in the following Azure regions. Note that the custom generative (document field extraction) model in AI Studio is only available in North Central US region:
East US
West US2
West Europe
North Central US
This content applies to:v4.0 (preview) | Previous versions:v3.1 (GA)
This content applies to:v3.1 (GA) | Latest version:v4.0 (preview)
Note
Add-on capabilities are available within all models except for the Business card model.
Capabilities
Document Intelligence supports more sophisticated and modular analysis capabilities. Use the add-on features to extend the results to include more features extracted from your documents. Some add-on features incur an extra cost. These optional features can be enabled and disabled depending on the scenario of the document extraction. To enable a feature, add the associated feature name to the features query string property. You can enable more than one add-on feature on a request by providing a comma-separated list of features. The following add-on capabilities are available for 2023-07-31 (GA) and later releases.
Not all add-on capabilities are supported by all models. For more information, seemodel data extraction.
Add-on capabilities are currently not supported for Microsoft Office file types.
Document Intelligence supports optional features that can be enabled and disabled depending on the document extraction scenario. The following add-on capabilities are available for 2023-10-31-preview, and later releases:
The query fields implementation in the 2023-10-30-preview API is different from the last preview release. The new implementation is less expensive and works well with structured documents.
✱ Add-On - Query fields are priced differently than the other add-on features. See pricing for details.
Supported file formats
PDF
Images: JPEG/JPG, PNG, BMP, TIFF, HEIF
✱ Microsoft Office files are currently not supported.
High resolution extraction
The task of recognizing small text from large-size documents, like engineering drawings, is a challenge. Often the text is mixed with other graphical elements and has varying fonts, sizes, and orientations. Moreover, the text can be broken into separate parts or connected with other symbols. Document Intelligence now supports extracting content from these types of documents with the ocr.highResolution capability. You get improved quality of content extraction from A1/A2/A3 documents by enabling this add-on capability.
# Analyze a document at a URL:
formUrl = "https://github.com/Azure-Samples/document-intelligence-code-samples/blob/main/Data/add-on/add-on-highres.png?raw=true"
poller = document_intelligence_client.begin_analyze_document(
"prebuilt-layout",
AnalyzeDocumentRequest(url_source=formUrl),
features=[DocumentAnalysisFeature.OCR_HIGH_RESOLUTION], # Specify which add-on capabilities to enable.
)
result: AnalyzeResult = poller.result()
# [START analyze_with_highres]
if result.styles and any([style.is_handwritten for style in result.styles]):
print("Document contains handwritten content")
else:
print("Document does not contain handwritten content")
for page in result.pages:
print(f"----Analyzing layout from page #{page.page_number}----")
print(f"Page has width: {page.width} and height: {page.height}, measured with unit: {page.unit}")
if page.lines:
for line_idx, line in enumerate(page.lines):
words = get_words(page, line)
print(
f"...Line # {line_idx} has word count {len(words)} and text '{line.content}' "
f"within bounding polygon '{line.polygon}'"
)
for word in words:
print(f"......Word '{word.content}' has a confidence of {word.confidence}")
if page.selection_marks:
for selection_mark in page.selection_marks:
print(
f"Selection mark is '{selection_mark.state}' within bounding polygon "
f"'{selection_mark.polygon}' and has a confidence of {selection_mark.confidence}"
)
if result.tables:
for table_idx, table in enumerate(result.tables):
print(f"Table # {table_idx} has {table.row_count} rows and " f"{table.column_count} columns")
if table.bounding_regions:
for region in table.bounding_regions:
print(f"Table # {table_idx} location on page: {region.page_number} is {region.polygon}")
for cell in table.cells:
print(f"...Cell[{cell.row_index}][{cell.column_index}] has text '{cell.content}'")
if cell.bounding_regions:
for region in cell.bounding_regions:
print(f"...content on page {region.page_number} is within bounding polygon '{region.polygon}'")
# Analyze a document at a URL:
url = "(https://github.com/Azure-Samples/document-intelligence-code-samples/blob/main/Data/add-on/add-on-highres.png?raw=true"
poller = document_analysis_client.begin_analyze_document_from_url(
"prebuilt-layout", document_url=url, features=[AnalysisFeature.OCR_HIGH_RESOLUTION] # Specify which add-on capabilities to enable.
)
result = poller.result()
# [START analyze_with_highres]
if any([style.is_handwritten for style in result.styles]):
print("Document contains handwritten content")
else:
print("Document does not contain handwritten content")
for page in result.pages:
print(f"----Analyzing layout from page #{page.page_number}----")
print(
f"Page has width: {page.width} and height: {page.height}, measured with unit: {page.unit}"
)
for line_idx, line in enumerate(page.lines):
words = line.get_words()
print(
f"...Line # {line_idx} has word count {len(words)} and text '{line.content}' "
f"within bounding polygon '{format_polygon(line.polygon)}'"
)
for word in words:
print(
f"......Word '{word.content}' has a confidence of {word.confidence}"
)
for selection_mark in page.selection_marks:
print(
f"Selection mark is '{selection_mark.state}' within bounding polygon "
f"'{format_polygon(selection_mark.polygon)}' and has a confidence of {selection_mark.confidence}"
)
for table_idx, table in enumerate(result.tables):
print(
f"Table # {table_idx} has {table.row_count} rows and "
f"{table.column_count} columns"
)
for region in table.bounding_regions:
print(
f"Table # {table_idx} location on page: {region.page_number} is {format_polygon(region.polygon)}"
)
for cell in table.cells:
print(
f"...Cell[{cell.row_index}][{cell.column_index}] has text '{cell.content}'"
)
for region in cell.bounding_regions:
print(
f"...content on page {region.page_number} is within bounding polygon '{format_polygon(region.polygon)}'"
)
The ocr.formula capability extracts all identified formulas, such as mathematical equations, in the formulas collection as a top level object under content. Inside content, detected formulas are represented as :formula:. Each entry in this collection represents a formula that includes the formula type as inline or display, and its LaTeX representation as value along with its polygon coordinates. Initially, formulas appear at the end of each page.
# Analyze a document at a URL:
formUrl = "https://github.com/Azure-Samples/document-intelligence-code-samples/blob/main/Data/add-on/layout-formulas.png?raw=true"
poller = document_intelligence_client.begin_analyze_document(
"prebuilt-layout",
AnalyzeDocumentRequest(url_source=formUrl),
features=[DocumentAnalysisFeature.FORMULAS], # Specify which add-on capabilities to enable
)
result: AnalyzeResult = poller.result()
# [START analyze_formulas]
for page in result.pages:
print(f"----Formulas detected from page #{page.page_number}----")
if page.formulas:
inline_formulas = [f for f in page.formulas if f.kind == "inline"]
display_formulas = [f for f in page.formulas if f.kind == "display"]
# To learn the detailed concept of "polygon" in the following content, visit: https://aka.ms/bounding-region
print(f"Detected {len(inline_formulas)} inline formulas.")
for formula_idx, formula in enumerate(inline_formulas):
print(f"- Inline #{formula_idx}: {formula.value}")
print(f" Confidence: {formula.confidence}")
print(f" Bounding regions: {formula.polygon}")
print(f"\nDetected {len(display_formulas)} display formulas.")
for formula_idx, formula in enumerate(display_formulas):
print(f"- Display #{formula_idx}: {formula.value}")
print(f" Confidence: {formula.confidence}")
print(f" Bounding regions: {formula.polygon}")
# Analyze a document at a URL:
url = "https://github.com/Azure-Samples/document-intelligence-code-samples/blob/main/Data/add-on/layout-formulas.png?raw=true"
poller = document_analysis_client.begin_analyze_document_from_url(
"prebuilt-layout", document_url=url, features=[AnalysisFeature.FORMULAS] # Specify which add-on capabilities to enable
)
result = poller.result()
# [START analyze_formulas]
for page in result.pages:
print(f"----Formulas detected from page #{page.page_number}----")
inline_formulas = [f for f in page.formulas if f.kind == "inline"]
display_formulas = [f for f in page.formulas if f.kind == "display"]
print(f"Detected {len(inline_formulas)} inline formulas.")
for formula_idx, formula in enumerate(inline_formulas):
print(f"- Inline #{formula_idx}: {formula.value}")
print(f" Confidence: {formula.confidence}")
print(f" Bounding regions: {format_polygon(formula.polygon)}")
print(f"\nDetected {len(display_formulas)} display formulas.")
for formula_idx, formula in enumerate(display_formulas):
print(f"- Display #{formula_idx}: {formula.value}")
print(f" Confidence: {formula.confidence}")
print(f" Bounding regions: {format_polygon(formula.polygon)}")
"content": ":formula:",
"pages": [
{
"pageNumber": 1,
"formulas": [
{
"kind": "inline",
"value": "\\frac { \\partial a } { \\partial b }",
"polygon": [...],
"span": {...},
"confidence": 0.99
},
{
"kind": "display",
"value": "y = a \\times b + a \\times c",
"polygon": [...],
"span": {...},
"confidence": 0.99
}
]
}
]
Font property extraction
The ocr.font capability extracts all font properties of text extracted in the styles collection as a top-level object under content. Each style object specifies a single font property, the text span it applies to, and its corresponding confidence score. The existing style property is extended with more font properties such as similarFontFamily for the font of the text, fontStyle for styles such as italic and normal, fontWeight for bold or normal, color for color of the text, and backgroundColor for color of the text bounding box.
# Analyze a document at a URL:
formUrl = "https://github.com/Azure-Samples/document-intelligence-code-samples/blob/main/Data/receipt/receipt-with-tips.png?raw=true"
poller = document_intelligence_client.begin_analyze_document(
"prebuilt-layout",
AnalyzeDocumentRequest(url_source=formUrl),
features=[DocumentAnalysisFeature.STYLE_FONT] # Specify which add-on capabilities to enable.
)
result: AnalyzeResult = poller.result()
# [START analyze_fonts]
# DocumentStyle has the following font related attributes:
similar_font_families = defaultdict(list) # e.g., 'Arial, sans-serif
font_styles = defaultdict(list) # e.g, 'italic'
font_weights = defaultdict(list) # e.g., 'bold'
font_colors = defaultdict(list) # in '#rrggbb' hexadecimal format
font_background_colors = defaultdict(list) # in '#rrggbb' hexadecimal format
if result.styles and any([style.is_handwritten for style in result.styles]):
print("Document contains handwritten content")
else:
print("Document does not contain handwritten content")
return
print("\n----Fonts styles detected in the document----")
# Iterate over the styles and group them by their font attributes.
for style in result.styles:
if style.similar_font_family:
similar_font_families[style.similar_font_family].append(style)
if style.font_style:
font_styles[style.font_style].append(style)
if style.font_weight:
font_weights[style.font_weight].append(style)
if style.color:
font_colors[style.color].append(style)
if style.background_color:
font_background_colors[style.background_color].append(style)
print(f"Detected {len(similar_font_families)} font families:")
for font_family, styles in similar_font_families.items():
print(f"- Font family: '{font_family}'")
print(f" Text: '{get_styled_text(styles, result.content)}'")
print(f"\nDetected {len(font_styles)} font styles:")
for font_style, styles in font_styles.items():
print(f"- Font style: '{font_style}'")
print(f" Text: '{get_styled_text(styles, result.content)}'")
print(f"\nDetected {len(font_weights)} font weights:")
for font_weight, styles in font_weights.items():
print(f"- Font weight: '{font_weight}'")
print(f" Text: '{get_styled_text(styles, result.content)}'")
print(f"\nDetected {len(font_colors)} font colors:")
for font_color, styles in font_colors.items():
print(f"- Font color: '{font_color}'")
print(f" Text: '{get_styled_text(styles, result.content)}'")
print(f"\nDetected {len(font_background_colors)} font background colors:")
for font_background_color, styles in font_background_colors.items():
print(f"- Font background color: '{font_background_color}'")
print(f" Text: '{get_styled_text(styles, result.content)}'")
# Analyze a document at a URL:
url = "https://github.com/Azure-Samples/document-intelligence-code-samples/blob/main/Data/receipt/receipt-with-tips.png?raw=true"
poller = document_analysis_client.begin_analyze_document_from_url(
"prebuilt-layout", document_url=url, features=[AnalysisFeature.STYLE_FONT] # Specify which add-on capabilities to enable.
)
result = poller.result()
# [START analyze_fonts]
# DocumentStyle has the following font related attributes:
similar_font_families = defaultdict(list) # e.g., 'Arial, sans-serif
font_styles = defaultdict(list) # e.g, 'italic'
font_weights = defaultdict(list) # e.g., 'bold'
font_colors = defaultdict(list) # in '#rrggbb' hexadecimal format
font_background_colors = defaultdict(list) # in '#rrggbb' hexadecimal format
if any([style.is_handwritten for style in result.styles]):
print("Document contains handwritten content")
else:
print("Document does not contain handwritten content")
print("\n----Fonts styles detected in the document----")
# Iterate over the styles and group them by their font attributes.
for style in result.styles:
if style.similar_font_family:
similar_font_families[style.similar_font_family].append(style)
if style.font_style:
font_styles[style.font_style].append(style)
if style.font_weight:
font_weights[style.font_weight].append(style)
if style.color:
font_colors[style.color].append(style)
if style.background_color:
font_background_colors[style.background_color].append(style)
print(f"Detected {len(similar_font_families)} font families:")
for font_family, styles in similar_font_families.items():
print(f"- Font family: '{font_family}'")
print(f" Text: '{get_styled_text(styles, result.content)}'")
print(f"\nDetected {len(font_styles)} font styles:")
for font_style, styles in font_styles.items():
print(f"- Font style: '{font_style}'")
print(f" Text: '{get_styled_text(styles, result.content)}'")
print(f"\nDetected {len(font_weights)} font weights:")
for font_weight, styles in font_weights.items():
print(f"- Font weight: '{font_weight}'")
print(f" Text: '{get_styled_text(styles, result.content)}'")
print(f"\nDetected {len(font_colors)} font colors:")
for font_color, styles in font_colors.items():
print(f"- Font color: '{font_color}'")
print(f" Text: '{get_styled_text(styles, result.content)}'")
print(f"\nDetected {len(font_background_colors)} font background colors:")
for font_background_color, styles in font_background_colors.items():
print(f"- Font background color: '{font_background_color}'")
print(f" Text: '{get_styled_text(styles, result.content)}'")
The ocr.barcode capability extracts all identified barcodes in the barcodes collection as a top level object under content. Inside the content, detected barcodes are represented as :barcode:. Each entry in this collection represents a barcode and includes the barcode type as kind and the embedded barcode content as value along with its polygon coordinates. Initially, barcodes appear at the end of each page. The confidence is hard-coded for as 1.
# Analyze a document at a URL:
url = "https://github.com/Azure-Samples/document-intelligence-code-samples/blob/main/Data/add-on/add-on-barcodes.jpg?raw=true"
poller = document_analysis_client.begin_analyze_document_from_url(
"prebuilt-layout", document_url=url, features=[AnalysisFeature.BARCODES] # Specify which add-on capabilities to enable.
)
result = poller.result()
# [START analyze_barcodes]
# Iterate over extracted barcodes on each page.
for page in result.pages:
print(f"----Barcodes detected from page #{page.page_number}----")
print(f"Detected {len(page.barcodes)} barcodes:")
for barcode_idx, barcode in enumerate(page.barcodes):
print(f"- Barcode #{barcode_idx}: {barcode.value}")
print(f" Kind: {barcode.kind}")
print(f" Confidence: {barcode.confidence}")
print(f" Bounding regions: {format_polygon(barcode.polygon)}")
Adding the languages feature to the analyzeResult request predicts the detected primary language for each text line along with the confidence in the languages collection under analyzeResult.
# Analyze a document at a URL:
formUrl = "https://github.com/Azure-Samples/document-intelligence-code-samples/blob/main/Data/add-on/add-on-fonts_and_languages.png?raw=true"
poller = document_intelligence_client.begin_analyze_document(
"prebuilt-layout",
AnalyzeDocumentRequest(url_source=formUrl),
features=[DocumentAnalysisFeature.LANGUAGES] # Specify which add-on capabilities to enable.
)
result: AnalyzeResult = poller.result()
# [START analyze_languages]
print("----Languages detected in the document----")
if result.languages:
print(f"Detected {len(result.languages)} languages:")
for lang_idx, lang in enumerate(result.languages):
print(f"- Language #{lang_idx}: locale '{lang.locale}'")
print(f" Confidence: {lang.confidence}")
print(
f" Text: '{','.join([result.content[span.offset : span.offset + span.length] for span in lang.spans])}'"
)
# Analyze a document at a URL:
url = "https://github.com/Azure-Samples/document-intelligence-code-samples/blob/main/Data/add-on/add-on-fonts_and_languages.png?raw=true"
poller = document_analysis_client.begin_analyze_document_from_url(
"prebuilt-layout", document_url=url, features=[AnalysisFeature.LANGUAGES] # Specify which add-on capabilities to enable.
)
result = poller.result()
# [START analyze_languages]
print("----Languages detected in the document----")
print(f"Detected {len(result.languages)} languages:")
for lang_idx, lang in enumerate(result.languages):
print(f"- Language #{lang_idx}: locale '{lang.locale}'")
print(f" Confidence: {lang.confidence}")
print(f" Text: '{','.join([result.content[span.offset : span.offset + span.length] for span in lang.spans])}'")
The searchable PDF capability enables you to convert an analog PDF, such as scanned-image PDF files, to a PDF with embedded text. The embedded text enables deep text search within the PDF's extracted content by overlaying the detected text entities on top of the image files.
Important
Currently, the searchable PDF capability is only supported by Read OCR model prebuilt-read. When using this feature, please specify the modelId as prebuilt-read, as other model types will return error for this preview version.
Searchable PDF is included with the 2024-07-31-preview prebuilt-read model with no usage cost for general PDF consumption.
Use searchable PDF
To use searchable PDF, make a POST request using the Analyze operation and specify the output format as pdf:
POST /documentModels/prebuilt-read:analyze?output=pdf
{...}
202
Once the Analyze operation is complete, make a GET request to retrieve the Analyze operation results.
Upon successful completion, the PDF can be retrieved and downloaded as application/pdf. This operation allows direct downloading of the embedded text form of PDF instead of Base64-encoded JSON.
// Monitor the operation until completion.
GET /documentModels/prebuilt-read/analyzeResults/{resultId}
200
{...}
// Upon successful completion, retrieve the PDF as application/pdf.
GET /documentModels/prebuilt-read/analyzeResults/{resultId}/pdf
200 OK
Content-Type: application/pdf
Key-value Pairs
In earlier API versions, the prebuilt-document model extracted key-value pairs from forms and documents. With the addition of the keyValuePairs feature to prebuilt-layout, the layout model now produces the same results.
Key-value pairs are specific spans within the document that identify a label or key and its associated response or value. In a structured form, these pairs could be the label and the value the user entered for that field. In an unstructured document, they could be the date a contract was executed on based on the text in a paragraph. The AI model is trained to extract identifiable keys and values based on a wide variety of document types, formats, and structures.
Keys can also exist in isolation when the model detects that a key exists, with no associated value or when processing optional fields. For example, a middle name field can be left blank on a form in some instances. Key-value pairs are spans of text contained in the document. For documents where the same value is described in different ways, for example, customer/user, the associated key is either customer or user (based on context).
Query fields are an add-on capability to extend the schema extracted from any prebuilt model or define a specific key name when the key name is variable. To use query fields, set the features to queryFields and provide a comma-separated list of field names in the queryFields property.
Document Intelligence now supports query field extractions. With query field extraction, you can add fields to the extraction process using a query request without the need for added training.
Use query fields when you need to extend the schema of a prebuilt or custom model or need to extract a few fields with the output of layout.
Query fields are a premium add-on capability. For best results, define the fields you want to extract using camel case or Pascal case field names for multi-word field names.
Query fields support a maximum of 20 fields per request. If the document contains a value for the field, the field and value are returned.
This release has a new implementation of the query fields capability that is priced lower than the earlier implementation and should be validated.
Note
Document Intelligence Studio query field extraction is currently available with the Layout and Prebuilt models 2024-02-29-preview2023-10-31-preview API and later releases except for the US tax models (W2, 1098s, and 1099s models).
Query field extraction
For query field extraction, specify the fields you want to extract and Document Intelligence analyzes the document accordingly. Here's an example:
If you're processing a contract in the Document Intelligence Studio, use the 2024-02-29-preview or 2023-10-31-preview versions:
You can pass a list of field labels like Party1, Party2, TermsOfUse, PaymentTerms, PaymentDate, and TermEndDate as part of the analyze document request.
Document Intelligence is able to analyze and extract the field data and return the values in a structured JSON output.
In addition to the query fields, the response includes text, tables, selection marks, and other relevant data.
# Analyze a document at a URL:
formUrl = "https://github.com/Azure-Samples/document-intelligence-code-samples/blob/main/Data/invoice/simple-invoice.png?raw=true"
poller = document_intelligence_client.begin_analyze_document(
"prebuilt-layout",
AnalyzeDocumentRequest(url_source=formUrl),
features=[DocumentAnalysisFeature.QUERY_FIELDS], # Specify which add-on capabilities to enable.
query_fields=["Address", "InvoiceNumber"], # Set the features and provide a comma-separated list of field names.
)
result: AnalyzeResult = poller.result()
print("Here are extra fields in result:\n")
if result.documents:
for doc in result.documents:
if doc.fields and doc.fields["Address"]:
print(f"Address: {doc.fields['Address'].value_string}")
if doc.fields and doc.fields["InvoiceNumber"]:
print(f"Invoice number: {doc.fields['InvoiceNumber'].value_string}")
Coming soon: Throughout 2024 we will be phasing out GitHub Issues as the feedback mechanism for content and replacing it with a new feedback system. For more information see: https://aka.ms/ContentUserFeedback.