AWS Textract

AWS Textract é um serviço poderoso de IA da Amazon Web Services projetado para extrair automaticamente texto impresso, manuscrito, tabelas, formulários, pares chave-valor e outros dados estruturados de documentos e imagens escaneados. O Textract usa reconhecimento óptico de caracteres (OCR) avançado e análise de documentos para transformar documentos em dados acionáveis, habilitando automação, analytics, compliance e mais.

Com o AWS Textract, você pode:

  • Extrair texto de imagens e documentos: Reconhecer texto impresso e manuscrito em formatos como PDF, JPEG, PNG ou TIFF
  • Detectar e extrair tabelas: Encontrar automaticamente tabelas e gerar seu conteúdo estruturado
  • Analisar formulários e pares chave-valor: Extrair dados estruturados de formulários, incluindo campos e seus valores correspondentes
  • Identificar assinaturas e recursos de layout: Detectar assinaturas, layout geométrico e relacionamentos entre elementos do documento
  • Personalizar a extração com queries: Extrair campos e respostas específicos usando extração baseada em queries (por exemplo, "Qual é o número da fatura?")

No Zoen, a integração com o AWS Textract capacita seus agentes a processar documentos de forma inteligente como parte dos workflows. Isso libera cenários de automação como entrada de dados de faturas, documentos de onboarding, contratos, recibos e mais. Seus agentes podem extrair dados relevantes, analisar formulários estruturados e gerar resumos ou relatórios diretamente de uploads ou URLs de documentos. Ao conectar o Zoen ao AWS Textract, você reduz esforço manual, melhora a precisão dos dados e agiliza processos de negócio com entendimento robusto de documentos.

Usage Instructions

Integrate AWS Textract into your workflow to extract text, tables, forms, and key-value pairs from documents. Single-page mode supports JPEG, PNG, and single-page PDF. Multi-page mode supports multi-page PDF and TIFF.

Actions

textract_parser

Input

ParameterTypeRequiredDescription
filefileNoDocument to be processed (JPEG, PNG, or single-page PDF).

Output

ParameterTypeDescription
blocksarrayArray of Block objects containing detected text, tables, forms, and other elements
BlockTypestringType of block (PAGE, LINE, WORD, TABLE, CELL, KEY_VALUE_SET, etc.)
IdstringUnique identifier for the block
TextstringThe text content (for LINE and WORD blocks)
TextTypestringType of text (PRINTED or HANDWRITING)
ConfidencenumberConfidence score (0-100)
PagenumberPage number
GeometryobjectLocation and bounding box information
BoundingBoxobjectBoundingBox output from the tool
HeightnumberHeight as ratio of document height
LeftnumberLeft position as ratio of document width
TopnumberTop position as ratio of document height
WidthnumberWidth as ratio of document width
PolygonarrayPolygon coordinates
XnumberX coordinate
YnumberY coordinate
RelationshipsarrayRelationships to other blocks
TypestringRelationship type (CHILD, VALUE, ANSWER, etc.)
IdsarrayIDs of related blocks
EntityTypesarrayEntity types for KEY_VALUE_SET (KEY or VALUE)
SelectionStatusstringFor checkboxes: SELECTED or NOT_SELECTED
RowIndexnumberRow index for table cells
ColumnIndexnumberColumn index for table cells
RowSpannumberRow span for merged cells
ColumnSpannumberColumn span for merged cells
QueryobjectQuery information for QUERY blocks
TextstringQuery text
AliasstringQuery alias
PagesarrayPages to search
documentMetadataobjectMetadata about the analyzed document
pagesnumberNumber of pages in the document
modelVersionstringVersion of the Textract model used for processing

textract_analyze_expense

Extract structured invoice and receipt fields using AWS Textract AnalyzeExpense

Input

ParameterTypeRequiredDescription
accessKeyIdstringYesAWS Access Key ID
secretAccessKeystringYesAWS Secret Access Key
regionstringYesAWS region for Textract service (e.g., us-east-1)
processingModestringNoDocument type: single-page or multi-page. Defaults to single-page.
filefileNoInvoice or receipt to be processed (JPEG, PNG, or single-page PDF).
filePathstringNoURL to an invoice or receipt to be processed, if not uploaded directly.
s3UristringNoS3 URI for multi-page processing (s3://bucket/key).

Output

ParameterTypeDescription
expenseDocumentsarrayDetected expense documents with summary fields and line items
expenseIndexnumberIndex of the expense document
summaryFieldsarrayHeader fields such as vendor name, invoice date, and totals
lineItemGroupsarrayGroups of line items (e.g., purchased items and their prices)
lineItemGroupIndexnumberIndex of the line item group
lineItemsarrayIndividual line items within the group
lineItemExpenseFieldsarrayFields for a single line item (description, quantity, price)
documentMetadataobjectMetadata about the analyzed document
pagesnumberNumber of pages in the document
modelVersionstringVersion of the AnalyzeExpense model used (multi-page/async only)

textract_analyze_id

Extract identity document fields using AWS Textract AnalyzeID

Input

ParameterTypeRequiredDescription
accessKeyIdstringYesAWS Access Key ID
secretAccessKeystringYesAWS Secret Access Key
regionstringYesAWS region for Textract service (e.g., us-east-1)
filefileNoFront of the identity document (JPEG, PNG, or PDF).
filePathstringNoURL to the front of the identity document, if not uploaded directly.
fileBackfileNoBack of the identity document, if applicable (JPEG, PNG, or PDF).
filePathBackstringNoURL to the back of the identity document, if not uploaded directly.

Output

ParameterTypeDescription
identityDocumentsarrayDetected identity documents with normalized fields
documentIndexnumberIndex of the document page set
identityDocumentFieldsarrayNormalized fields such as FIRST_NAME, LAST_NAME, DATE_OF_BIRTH, DOCUMENT_NUMBER, EXPIRATION_DATE
typeobjectNormalized field label
textstringField label text
confidencenumberConfidence score (0-100)
valueDetectionobjectDetected value for the field, with a normalized value for dates
textstringField value text
confidencenumberConfidence score (0-100)
documentMetadataobjectMetadata about the analyzed document
pagesnumberNumber of pages analyzed
modelVersionstringVersion of the AnalyzeID model used for processing

On this page