Configure OCR Bridge step and OCR providers

The OCR Bridge step lets you process documents in a Business Process (BP) and pass optical character recognition (OCR) results to the next step, such as a Model or labelling Manual Task step.

Add no-code OCR Bridge step to Business Process

To add a no-code OCR Bridge step to a BP, follow the steps below:

1. In an existing BP, navigate to the Workflow tab.

   

2. In the step panel on the right, select Bot and click No-Code Bridge to open the list of available no-code steps.

3. Add the No-Code OCR Bridge step to the Workflow Designer canvas in one of the following ways:

   • Reuse an existing step. Drag an existing No-Code OCR Bridge step from the panel on the right to the canvas. In this case, any changes you make to the step are applied across all BPs where the same step is included.

     

   • Copy an existing step. In the panel on the right, hover over the No-Code OCR Bridge step and click the displayed Copy button. As a result, you create an independent No-Code OCR Bridge step instance: the changes you make in the copy are not synchronized with the source.

4. Double-click the step on the canvas and set a name for the step in the Name field.

   

5. Navigate to the Input / Output tab of the OCR step to set up the input and output contracts.

   Note that if you set up the inputs and outputs, the whole BP becomes a schema-based one, and you will have to configure inputs and outputs for the rest of the steps in the same BP (if those are not configured yet). For details, see the guide.

6. Go back to the Design tab and configure the OCR step parameters.

7. Click Save to save the created OCR configuration.

Alternatively, you can use a regular OCR Bridge step to set up the OCR processing in a BP. The setup sequence is similar to the one described above.

Configure OCR step

To configure an OCR step, double-click the step in the canvas to open the configuration window (if it is not open yet) and perform the following actions on the Design tab:

1. Set the OCR step input parameters.

2. Choose an OCR engine.

3. Set the OCR parameters corresponding to the chosen OCR engine.

Set step input parameters

Parameter	Code	Default value	Description
Document URL	document_url	original_document_url	URL of the document to be processed by OCR
OCR Type	ocr_type	Enriched	OCR type values: Standard and Enriched (prepare output for a labelling Manual Task)
S3 bucket name	s3_bucket_name	doc-upload	S3 bucket to store results
S3 folder name	s3_folder_name	empty	S3 folder to store results
OCR Provider	ocr_provider	ABBYY	OCR provider: ABBYY Google Vision Azure Form Recognition
OCR parameters	ocr_parameters	See OCR JSON sample	See the OCR parameters.

Choose OCR provider

To configure an OCR provider, choose one from the OCR Provider drop-down list and set the required paratemers. For details, read Set OCR parameters.

Work.AI supports the following OCR providers:

• ABBYY
• Azure AI Document Intelligence
• Google Vision

For providers other than ABBYY, mind the following considerations:

• If you select Google Vision or Azure AI Document Intelligence, check that the default configuration includes the following parameters:

  • In the OCR Type field, choose Enriched.

  • In the OCR parameters json box, update the settings as follows:

    • exportFormat: xmlForCorrectedImage,xmlWithoutRecognitionVariants
    • writeRecognitionVariants: true
    • skipPreprocessing: false
    • storePreprocessedDoc: true
    • timeout: 60

    Note: if any of these settings are not configured as listed, OCR throws an exception.

    

• To use Google Vision or Azure AI Document Intelligence, set up additional Secrets Vault credentials.

Set Secrets Vault credentials

The Work.AI platform does not manage Google secret keys and Azure credentials, so you must add them using one of the methods described below.

Method 1. New installation

As you install Work.AI, set the following properties in config.yml:

• ocr_google_key
• ocr_azure_password
• ocr_azure_endpoint

note

ocr_google_key must contain a Base64-encoded JSON Google key. Otherwise, escaping issues can occur in Ansible. If you pass the key as JSON or as separate properties, the private key can cause issues because it contains \n new lines and vault values with new lines.

Method 2. Existing installation

Set or create the same properties in config_overrides.yml in /opt/workfusion/wf_installer.

To override the parameters in the backup, run:

./install.sh install installer-backup -e @config_overrides.yml

To restart the OCR service with the parameters from config_overrides.yml, run:

./install.sh configure ocr -e @config_overrides.yml

If the steps above cause errors, make sure that the Ansible vault pass and key properties are set:

cd /opt/workfusion/wf_installer

export ANSIBLE_VAULT_PASS=$(cat config_overrides.yml | grep ansible_vault_password | cut -c 25- | tail -1)

export ANSIBLE_PRIVATE_KEY_FILE="id_rsa"

Set OCR parameters

Parameter	Values	Description
exportFormat		OCR API export format. You can define multiple export formats using a comma as a separator: xml,xmlForCorrectedImage. Mind that, for correct processing of a labelling Manual Tasks, you cannot use more than three export formats.
	txt	Default export format.
	xml	Contains characters or words along with their location in the original document (coordinates or frames).
	xmlWithoutRecognitionVariants	The same as xml but does not contain character-related data.
	xmlForCorrectedImage	The same as xml, except the location is taken from a processed or adjusted document.
	pdfSearchable
	html
language	English German so on	Predefined language. You can also define multiple languages using a comma as a separator: English,German,Polish.
correctSkew	boolean	The page skew is detected and automatically corrected.
correctOrientation	boolean	The page orientation is detected. If it differs from normal, it is rotated automatically.
writeRecognitionVariants	boolean	Causes the xml and xmlForCorrectedImage formats to contain all variants of characters or words that OCR considers a possible recognition result.
invertImage	boolean	Inverts images.
discardColorImage	boolean	Leaves only the black-and-white plane in a prepared image.
skipPreprocessing	boolean	Disables all preprocessing steps for a document. The default value is false. If you set to true, the convertTo and changeDPI parameters are disabled.
useOnlyCustomRegions	boolean	Skips the original analyzing stage and extracts information from custom regions only.
useDefaultPattern	boolean	If true, it requires applying the default pattern from the OCR application bundle.
profile	documentConversion	Converts documents into editable formats.
	textExtraction	Extracts text from documents.
	barcodeRecognition	Extracts barcodes.
removeGarbageSize	integer	Removes garbage from images: excess dots smaller than a specific size. Optional; valid value > 0 and -1 for the automatically detected garbage size.
lowResolutionMode	boolean	Improves the recognition of images with low resolution.
enhanceLocalContrast	boolean	Specifies whether the local contrast of an image should be increased.
convertTo	none	Automatically detects the file type and conversion to TIFF before processing (only conversion to TIFF is supported). Accepts images: PDF, PNG, JPG, JPEG.
changeDPI	integer [50 : 3200] valid range null or 0 for disable	Specifies a new value for the resolution, for example, changeDPI=300. The recommended resolution for a source image is 300 DPI for typical texts (10 pt or larger) and 400-600 DPI for texts in smaller fonts (9 pt or smaller).
version	V_2	ABBYY FRE 12
	V_1	ABBYY FRE 11
storePreprocessedDoc	boolean	Specifies whether preprocessed documents should be saved in the data storage.
textTypes	normal typewriter matrix ocrA ocrB e13b	Specifies the type of the text in the document. You can define several text types using a comma as a separator: normal,matrix.
timeout	integer	Sets the maximum allowed timeout in seconds for a single task execution by an OCR Worker. The timeout countdown starts when an OCR Worker starts executing a task. When the task execution exceeds the given timeout, the task is forcibly killed. If you do not set the value, the default value (0) is used, and your Business Process (BP) does not have time to process a task.

If native_pdf_processing is set to true, the skipPreprocessing and convertTo parameters are disabled. Such parameters as changeDPI and storePreprocessedDoc take effect only on saving pages of an input PDF.

troubleshooting

In case of issues with input data, see the Issues related to input data support guide.

OCR JSON sample

{
    "exportFormat":"xmlForCorrectedImage,xmlWithoutRecognitionVariants",
    "language": "English",
    "correctSkew": true,
    "correctOrientation": true,
    "writeRecognitionVariants": true,
    "invertImage": false,
    "discardColorImage": false,
    "skipPreprocessing": false,
    "useOnlyCustomRegions": false,
    "useDefaultPattern": false,
    "profile": "documentConversion",
    "removeGarbageSize": 0,
    "lowResolutionMode": false,
    "enhanceLocalContrast": false,
    "allowedRegionTypes": null,
    "changeDPI": 0,
    "version": "V_2",
    "storePreprocessedDoc": true,
    "textTypes": ["normal"],
    "timeout": 60
}

Review output

• OCR Result: document processing results in the JSON format

  • Code: ocr_result

  • Example:

  {
      "originalDocument": "https://minio-host/doc-upload/1.pdf",
      "documents": {
          "xmlForCorrectedImage": "https://minio-host/doc-upload/bp-id/task-id/1- xmlForCorrectedImage.xml",
          "xmlWithoutRecognitionVariants": "https://minio-host/doc-upload/bp-id/task-id/1-xmlWithoutRecognitionVariants.xml"
      },
      "pageDetails": "https://minio-host/doc-upload/bp-id/task-id/preprocessed/links.json"
  }

• Task ID: the ID of an OCR task

  • Code: task_id
  • Example: 066416ed-20e8-4e96-9df1-7f7e133f1211

• Meta Info JSON extended info for the ToD Manual Task

  • Code: meta_info_json
  • Example:

  {
      "pages": "https://minio-host/doc-upload/bp-id/task-id/preprocessed/links.json",
      "ocrXmlUrl": "https://minio-host/doc-upload/bp-id/task-id/1-xmlWithoutRecognitionVariants.xml",
      "imageUrl": "https://minio-host/doc-upload/1.pdf"
  }

• Process time: record process time in milliseconds

  • Code: _sys_ocr_process_time
  • Example: 24053

See a JSON example for https://minio-host/doc-upload/bp-id/task-id/preprocessed/links.json below:

[
    {
        "imgUrl": "https://minio-host/doc-upload/bp-id/task-id/preprocessed/1.png",
        "charsUrl": "https://minio-host/doc-upload/bp-id/task-id/preprocessed/1.json",
        "height": 3300,
        "width": 2550,
        "charsCount": 2216
    }
]

tip

To set your Java Native Worker to work with OCR Bridge step data, it is recommended to use the respective dependency from the WorkFusion's repository:

<groupId>com.workfusion.spa</groupId>
<artifactId>bridge-step-contracts</artifactId>

Using the library facilitates the implementation and reduces the time to production for such solutions.

Set up OCR REST configuration properties

Property	Default value	Description
execution.bridge.task.queue	exec.<CT_HOST>.result.ocr	Name of the OCR Bridge queue
execution.bridge.prefetch.count	100	Prefetch count of the Bridge queue
execution.bridge.thread.count	20	Count of threads for consuming messages of the Bridge queue