Enterprise Document OCR (Optical Character Recognition)
Description
Identify and extract text in different types of documents.
This processor allows you to identify and extract text, including handwritten text, from documents in more than 200 languages. The processor also uses machine learning to perform a quality assessment of a document based on the readability of its content.
Extract general key-value pairs (entity and checkbox), tables, and generic entities from documents in addition to OCR text.
This processor applies advanced machine learning technologies to extract key-value pairs, checkboxes, and tables from documents more than 200 languages. This processor also leverages deep learning models to extract 11 generic entities that are common in various document types.
Extracts document content elements (text, tables, and lists) and creates context-aware chunks.
Layout Parser extracts document content elements like text, tables, and lists, and creates context-aware chunks that facilitate information retrieval in generative AI and discovery applications.
If a page of a multi-page input file is the correct document type and one of the supported versions, the processor performs entity extraction on the first supported document. If the processor doesn't find any applicable documents in the input file, the processor returns an error message.
Supported languages
Language Name
BCP 47 Tag
Script
Handwriting supported
English
en
Latn
Processor versions
Version ID
Release Channel
Description
pretrained-bankstatement-v1.0-2021-08-08
Stable
pretrained-bankstatement-v1.1-2021-08-13
Stable
pretrained-bankstatement-v2.0-2021-12-10
Stable
pretrained-bankstatement-v3.0-2022-05-16
Stable
This version assumes that the input file contains a single bank statement. Unlike the default version, this version does not check the input file for bank statements and will not return an error if no bank statements are found.
If a page of a multi-page input file is the correct document type and one of the supported versions, the processor performs entity extraction on the first supported document. If the processor doesn't find any applicable documents in the input file, the processor returns an error message.
Supported languages
Language Name
BCP 47 Tag
Script
Handwriting supported
English
en
Latn
Supported form/versions
2020 (standard and customized versions)
2019 (standard and customized versions)
2018 (standard and customized versions)
Processor versions
Version ID
Release Channel
Additional fields detected
Description
pretrained-w2-v1.0-2020-10-01
Stable
None
pretrained-w2-v1.1-2022-01-27
Stable
None
pretrained-w2-v1.2-2022-01-28
Stable
Show fields
AllocatedTips
ControlNumber
DependentCareBenefits
EIN
EmployeeAddress
EmployeeName
EmployerNameAndAddress
EmployerStateIdNumber_Line1
FederalIncomeTaxWithheld
FormYear
LocalIncomeTax_Line1
LocalityName_Line1
LocalWagesTipsEtc_Line1
MedicareTaxWithheld
MedicareWagesAndTips
NonqualifiedPlans
SocialSecurityTaxWithheld
SocialSecurityTips
SocialSecurityWages
SSN
State_Line1
StateIncomeTax_Line1
StateWagesTipsEtc_Line1
WagesTipsOtherCompensation
Quality improvements and supporting new fields; does not include splitter.
Quality improvements and support for box 12 fields and fine-grained predictions of EmployeeName, EmployeeAddress, and EmployerNameAndAddress, all of which are no longer part of the output and are replaced with additional fields.
The Online Duplicate Detection feature is currently processed in US data centers. Regional and multi-regional support is unavailable for this feature outside of the US.
This processor is supported by algorithms that are updated more frequently than new processor versions are released. For this reason, the processor might return different outputs over time even when using the same processor version. For example, the Online Duplicate Detection system monitors images present on the web. The system's behavior can then change more quickly than can be tracked in processor versions.
Refer to notes on Responsible AI[†] and Human review.[‡]
Supported languages
Language Name
BCP 47 Tag
Script
Handwriting supported
English
en
Latn
Supported form/versions
Support for US passports, passcards and driver's licenses.
If the multi-page input document contains more than one valid pay slips, the processor extracts entities from only the first valid pay slip. If no pay slips are found in the input file, the processor returns an error message.
Supported languages
Language Name
BCP 47 Tag
Script
Handwriting supported
English
en
Latn
Processor versions
Version ID
Release Channel
Additional fields detected
Description
pretrained-paystub-v1.0-2021-03-19
Stable
None
pretrained-paystub-v1.1-2021-08-13
Stable
Show fields
net_pay
net_pay_ytd
employee_account_number
Quality improvement and new fields support;
pretrained-paystub-v1.2-2021-12-10
Stable
None
pretrained-paystub-v2.0-2022-05-17
Release Candidate
Show fields
deduction_item
deduction_item/deduction_type
deduction_item/deduction_this_period
deduction_item/deduction_ytd
direct_deposit_item
direct_deposit_item/direct_deposit
direct_deposit_item/employee_account_number
earning_item
earning_item/earning_type
earning_item/earning_rate
earning_item/earning_hours
earning_item/earning_this_period
earning_item/earning_ytd
page_number
tax_item
tax_item/tax_type
tax_item/tax_this_period
tax_item/tax_ytd
federal_additional_tax
federal_allowance
federal_marital_status
state_additional_tax
state_allowance
state_marital_status
This version assumes that the input file contains a single pay slip. Unlike the default version, this version does not check the input file for pay slips and will not return an error if no pay slips are found.
Quality improvement, new fields support and new schema. Bonus, Commissions, Holiday, Overtime, Regular Pay and Vacation are now part of earning_item/earning_this_period, and their year-to-date versions are in earning_item/earning_ytd. Direct Deposit and Employee Account Number are now nested under direct_deposit_item.
Extract text and values from invoices such as invoice number, supplier name, invoice amount, tax amount, invoice date, due date.
The invoice Parser extracts both header and line item fields, such as invoice number, supplier name, invoice amount, tax amount, invoice date, due date, and line item amounts.
[*] This processor is only available to limited-access customers.
To request API access, fill out and submit the
Document AI limited
access customer request form.
The form requests information about you, your company, and your use case.
Note that a Google Cloud Project ID is required for access.
To create a new Google Cloud project, or identify your existing project's
Project ID see the following
instructions.
After you submit the form, the Document AI team will
review your request to ensure you meet the criteria for access.
If approved, you will receive an email with instructions on how to access
and use this feature.
[†]
Identity Document Proofing works to extract and evaluate information from ID documents that contributes to identifying whether the input image represents an authentic ID.
At Google Cloud, we prioritize helping customers safely develop and implement AI solutions, and Identity Proofing has been developed in accordance with Google's AI Principles.
Based on Google's AI Principles and current product design, we strongly recommend using caution and carefully evaluating the potential benefits and risks of using Identity Document Proofing for the following:
Decision-making without a human in the loop for predictions that can impact human rights.
In sensitive domains including but not limited to employment, access to public services, healthcare, and safety-critical contexts.
[‡] Always use Identity Proofing as part of your broader identity-detection process and workflow.
It is important that you have a human reviewer in your workflow to verify whether the predicted signals are accurate. The Identity Proofing processor isn't meant to replace human review of IDs in a workflow, but rather to assist human reviewers in validating ID documents. The Identity Proofing processor shouldn't be used as an automated decision tool to determine whether an ID is valid. With human review, customers can achieve higher document processing accuracy and help businesses evaluate predictions using purpose-built tools to enable those reviews.
Make sure that you review regulations in the region where you are implementing this technology, and research existing industry guidance to learn about policy guidelines and common fairness issues. Read about fairness in machine learning, including ways to mitigate bias in training datasets, evaluate your custom models for disparities in performance, and other considerations as you use your custom model.
We encourage customers to keep fairness, interpretability, and privacy and security best practices in mind when implementing Identity Proofing. To learn more about how to implement responsible AI, read Google's recommendations for Responsible AI practices.
[[["Easy to understand","easyToUnderstand","thumb-up"],["Solved my problem","solvedMyProblem","thumb-up"],["Other","otherUp","thumb-up"]],[["Hard to understand","hardToUnderstand","thumb-down"],["Incorrect information or sample code","incorrectInformationOrSampleCode","thumb-down"],["Missing the information/samples I need","missingTheInformationSamplesINeed","thumb-down"],["Other","otherDown","thumb-down"]],["Last updated 2025-07-08 UTC."],[[["Document AI offers various processors for digitizing, extracting, classifying, and summarizing document content, including Enterprise Document OCR for text recognition in over 200 languages."],["Custom Extractor uses generative AI or custom models to extract fields from documents, supporting multiple languages and specific data types, such as `dateTime`, `currency`, `number`, and has different versions with varying functionalities and lifespans."],["Form Parser extracts key-value pairs, checkboxes, tables, and entities from documents in over 200 languages, utilizing machine learning and deep learning techniques, and is capable of extracting fields such as `email`, `phone`, `url`, `date_time`, and more."],["Layout Parser, available for files such as PDF, HTML and DOCX, extracts content elements like text, tables and lists, supports a wide array of languages, and allows for processing up to 15 pages online or 500 in batch."],["Specialized pretrained processors like Bank Statement Parser, W2 Parser, US Passport Parser, Utility Parser, Identity Document Proofing Parser, Pay Slip Parser, US Driver License Parser, Expense Parser, and Invoice Parser are available for extracting specific information from different document types."]]],[]]
