DANIEL

Description : 

An OCR transcribes a written document seen as a picture into a numerical file, making it easier to manipulate. However, it may be interesting to go beyond this first process by obtaining a rich numercial text where named entities have a specific label. This is what DANIEL does : extracting the named entities from a written document.

How does it work : 

• Acquisition of the written document as an image
• Scanning of the document to extract its text
• This text is then analysed to detect the named entities
• A label is associated to each named entity 

DANIEL is an end-to-end software performing handwritten text recognition and named entity recognition on full-page documents. It is working with a fully convolutional encoder so it is able to deal with images of any size and uses an attention network with a LLM to extract named entities.

Applications : 

• Creation of databases linking entities within documents
• Analysing historical document
• Searching through documents for a specific named entity

• State of the art results in text recognition and named entity extraction
• Works with multiple languages
• Faster than other solutions
• End-to-end architecture