Data extraction

Pulling name, date of birth, and number off a document into your system.

After a person photographs an ID, data extraction turns the picture into clean fields that a system can search, compare, and store.

Field-level parsing and normalization of document data with per-field confidence scoring. Parsing splits the raw text into named fields, normalization puts values into a consistent format, and a confidence score shows how sure the system is about each field.

Data extraction typically follows these steps:

1. The system reads the text from the document, using OCR or a vision-language model.
2. It assigns the text to named fields, such as name, date of birth, and document number.
3. It normalizes the values, such as converting every date to one format.
4. It scores its confidence in each field.

Several kinds of cleanup happen along the way:

• Date formats, since countries write dates in different orders.
• Name handling, such as the order of given and family names and the treatment of accents.
• Cross-checks, such as comparing the printed name with the machine-readable zone.

Data extraction sits next to terms that are easy to mix up:

• Optical character recognition is the reading step, while data extraction covers the whole process.
• Manual review is where low-confidence cases go for a human decision.
• Case management is the workspace where reviewers handle those cases.