The purpose of IRI data masking tools is to discover and obfuscate Personally Identifiable Information (PII) in various data sources. Depending on the type of data source, we recommend different products in the IRI Data Protector Suite, typically IRI FieldShield or IRI DarkShield. 

Ultimately, the biggest challenge in protecting data is not the masking of the data, but reliably finding the PII in the source(s). This is particularly true for unstructured sources of data – like free text files, documents, and images – that DarkShield supports.

Handwritten text (like signatures) can be particularly difficult to work with. Parsing and accurately translating handwriting into digital equivalents requires a sophisticated AI-powered Optical Character Recognition solution. This is because handwriting varies widely in form and quality from person to person.

Collaboration between IRI and DeepLobe

IRI DarkShield 1.6.5 in IRI Workbench 11.0.1 now supports the discovery and masking of handwritten signatures in files. This new functionality is the result of the latest collaboration between IRI and DeepLobe. 

The combined solution, announced here and described in this article, is designed to address the challenge of detecting and masking PII in the form of signatures. When added to DarkShield, the DeepLobe AI model enables you to find and redact one or more signature images in a document (e.g., Word, PDF) or standalone image file.

Note that the signature detection functionality is not to be confused with another AI task called Handwritten Text Recognition. That is a more substantial challenge and is currently being developed for similar use in the on-premise DarkShield data masking tool. 

Signature Solution Installation and Setup

To get started, you will need:  

1. IRI DarkShield software, installed on Linux or Windows with Voracity
2. A valid IRI license key, created after DarkShield installation and registration
3. The DeepLobe signature model Docker container 
4. A DeepLobe API key, provided by IRI for evaluation or licensing purposes.

IRI Workbench and Signature Detection Feature

The front-end of the new signature detection option using DarkShield has now been added to IRI Workbench and is configured from the Data Class Library as a Data Matcher. For more information on what Data Matchers are, please read this article. 

Screenshot of IRI Data Class and Rule Library with a signature data class

To create a data matcher, you must open the Data Class Library and either create a new Data Class or access an existing Data Class. Clicking on a Data Class displays information about its Location Matchers and Data Matchers. 

It is in this section where you can then add a Data Matcher:

Dialog page for creating a Signature Matcher

Select from the dropdown box “Signature Detection” to specify the type of the Data Matcher. A data matcher for signature detection takes three arguments:

1. The listener URL, for HTTP/S requests to the on-premise DeepLobe container
2. The API key, to authenticate your license to use the DeepLobe signature model 
3. The confidence threshold, or the minimum percentage (score) of confidence that the signature detection model must meet for each location prediction to not be considered a false positive.

Once a Data Matcher is added to a Data Class, simply include the Data Class during the creation of a DarkShield job. 

To create your DarkShield File Search/Masking Job, launch the wizard from this menu: 

Screenshot of DarkShield Wizards drop-down menu

For more information about creating DarkShield Jobs please read this article. The first step will be to name your DarkShield job; which gets saved as a .dsc (DarkShield configuration) job file after it is built in the wizard (or edited in the GUI). 

Proceed through the next steps in the wizard as you normally would, but make sure you include the Data Class assigned with your Signature data matcher when prompted:

Screenshot of data class selection wizard page with Signature data class selected

Once the DarkShield Job (.dsc file) is created, you can review/modify your data source and targets, and reassign data classes if/as needed:

User Interface for reviewing and modifying .dsc configuration file

To run the DarkShield job from IRI Workbench, right-click on your .dsc file, then click Run As > IRI Search and Masking Job:

Screenshot showing how to run a DarkShield job using .dsc file

Warning: prior to running the job make sure the DarkShield API http(s) server is running and listening on the correct port. Furthermore, ensure DeepLobe’s docker image has been spun up and is reachable via http(s).

Sample File w/Signature Masked

The original JPG file of a check contains a handwritten signature in the bottom right corner.

The signature discovered in the check is redacted using masking.

What Comes Next

IRI is aware of the need and value of supporting the ability to parse handwriting and mask any PII within it. As such, as of 2025, IRI and DeepLobe are working together on the future goal of recognizing and masking PII in handwritten text generally.

With this upcoming feature, DarkShield should, in theory, be able to support the sanitization of PII in various new sources, such as doctor’s notes, customer order forms, etc. If you have any questions or use case requirements to share, please contact us at darkshield@iri.com.