How to process identity documents automatically

The following article explains how to process identity documents automatically with MyBiros, specifically focusing on automating the extraction of personal information. This service is valuable in various application scenarios, such as customer onboarding, both remotely and in person. In these cases, automating identity document processing can significantly streamline the customer registration process.

Description of the use case

Identity documents (such as identity cards, passports, driving licenses, and health cards) are categorized as structured documents. Therefore, it is possible to process them using traditional solutions based on rules and templates applied to the output of an OCR engine. However, several complications prevent traditional methods from fully automating this use case:

• Document acquisition is not always standardized: Identity documents may be scanned or photographed in rotated or low-quality formats, making them difficult to read. These issues pose challenges for rule-based and template-based approaches, which are unable to handle such variability effectively.
• Diversity of identity documents: While an individual identity document might be straightforward to interpret, each country has one or more types of identity documents. For a global use case, this means hundreds of different document formats. Traditional solutions would require creating separate rules and templates for each one.
• Document changes over time: As identity document formats evolve, additional rules and templates are needed to accommodate these changes.

These challenges can be overcome with innovative solutions that leverage Deep Learning techniques, eliminating the need for rules and templates. myBiros effectively addresses these issues by automating the process. For simplicity, the article will focus on the use case of the Italian identity card.

This version improves clarity, structure, and flow while maintaining the original meaning.

Building a use case with myBiros

MyBiros leverages Deep Learning techniques to eliminate the need for rules and templates, relying instead on a fully data-driven approach. Unlike traditional methods that focus solely on field positioning for extraction, myBiros uses semantic analysis, document geometry analysis, and layout interpretation to process documents effectively.

Creating a use case with MyBiros is straightforward and involves the following steps:

1. Collection of documents
2. Data annotation
3. Training
4. Service release and performance testing

‍

Collecting documents

The first step is to collect a small sample of reference documents, such as 10 Italian identity cards. This document collection is essential because Artificial Intelligence algorithms require training data to learn and develop the ability to extract information accurately. These reference documents serve as the foundation for training the algorithm to recognize and process similar documents in the future.

‍

Data annotation

The annotation phase transforms documents into data that can be understood by AI. With MyBiros' intuitive no-code interface, users can easily specify the information of interest by simply clicking on the data to be extracted. Additionally, MyBiros' AI helps accelerate the process by suggesting relevant information to extract, further streamlining the annotation phase.

‍

‍

Training

In this phase, the algorithm learns from the information prepared during the annotation phase. This step is fully automated, and within a few hours, you'll have a newly trained model. A key feature of the MyBiros platform is the ability to choose between training the algorithm from scratch or using one of MyBiros' pre-trained models from other domains to speed up training and improve accuracy.

During the training process, you can monitor model evaluation metrics, ensuring the accuracy of the extracted data. This allows for real-time insight into the model’s performance as it evolves.

‍

‍

Service delivery and performance testing

Once training is complete, the service can be tested through an intuitive interface that displays the results, allowing you to evaluate the model's performance. This interface also provides access to the new API associated with the created use case, along with examples of possible integrations, making it easy to implement and assess the effectiveness of the model in real-world scenarios.

‍

‍

Results and benefits

The result is an Artificial Intelligence model capable of understanding and extracting the specified information of interest from documents, encapsulated in an API that can be easily accessed remotely.

The benefits of implementing such a use case include:

• Reduction of errors caused by manual data entry
• Elimination of repetitive and tedious data entry tasks
• Time and cost savings
• More accurate and secure data extraction
• Faster and more efficient processes, making workflows more streamlined and immediate

These advantages provide significant improvements in operational efficiency and data handling.

Want to learn more about our solutions? Contact us, we are ready to assist you!