Extract data from a document Dbrain OCR

Extract data from a document Version 3 (Python)

Action group: Text recognition

Description

The action extracts data from documents (Passport 2-3 pages, Passport 5 pages, personal insurance policy number) and saves it as a dictionary containing extracted document data and an image with blocks from which the data was extracted. The action uses Dbrain services for data extraction, so it requires a vendor API key.

Action icon

Parameters and their settings

Property	Description	Type	Filling example	Mandatory field
Parameters
File path	The path to the file to extract the data from. Supported formats: jpg, jpeg, bmp, png.	Robin.FilePath		Yes
API key	A unique identifier for accessing the service.	Robin.String		Yes
Cloud server	If the value is «true», the action will send a request to the DBrain cloud server. If «false», the action will send a request to the local DBrain server.	Robin.Boolean	true	No
Document type	The type of document to extract data from.	Robin.String		Yes
Folder path	The path to the folder where a copy of the source file with superimposed blocks will be saved, into which the recognition action divides it. To save it, you also need to fill in the «File name» field.	Robin.FolderPath	C:\doc\img	No
File name	The name of the copy of the source file with superimposed blocks (without extension). The image will be created with the *.png extension. To save it, you also need to fill in the «Folder path» field.	Robin.String		No
Time out	The time in milliseconds during which data will be extracted from the document. The default value is 120000 ms.	Robin.Numeric		No
Overwrite	If «true», and a file with the same name and extension exists in the specified folder, then a new file with blocks will overwrite it. If «false», the file will not be overwritten, and the action will return an error.	Robin.Boolean	true	No
Results
Extracted text	The data of each field in the document extracted from the original image.	Robin.Dictionary
Image with blocks	The path to the image file with superimposed blocks.	Robin.FilePath
Recognition confidence	The accuracy of recognition of each field in the image is in the range from 0 to 1. The key is the name of the field, the value is the accuracy of recognition of this field.	Robin.Dictionary

Special conditions of use

Connecting to Dbrain: https://doc.dbrain.io/podklyuchenie/podklyuchenie-k-oblaku

Neural network recognizes only Russian documents.

For each document type, a set of fields that the robot will search for is defined. If the image does not contain the fields it is looking for, the field value will be returned empty. Keys are returned in Russian.

Passport keys (2-3 pp):

IssuedBy
IssuedDate
IssuedCode
Signature
LName,
FName,
MName,
Sex,
Photo
BirthDate
BirthPlace
MRZ,
Number

Registration keys (5 p):

Полный адрес
Дата регистрации
Регион
Район
Пункт
Р-н
Улица
Дом
Строение
Квартира
Подразделение
Код подразделения

Keys for SNILS:

Number
LName
FName
MName
BirthDate
BirthPlace
Sex
RegDate

The robot will return an error if:

The "Folder path" field is filled in and the "File name" field is not filled in.
the "File name" field is filled in and the "File path" is not filled in.
an invalid input file format is entered in the "File path" field.
there is already a file with the specified name on the specified path and the "overwrite" field = false.
an invalid API key is specified.
Timeout expired, no result was received.
Cloud server is not checked, local server is not deployed by the user on the host.

Example of use

Task

Recognize fields with document data from a file.

A document to recognize:

Solution

Use the "Extract data from a document" action.

Implementation

Move the "Extract data from a document" action to the workspace.
Set the parameters of the "Extract data from a document" action.
Click on the "Start" button in the top panel.

Result

The program robot completed successfully. The data fields have been recognized.

	Recognition accuracy
Номер СНИЛС	1
Фамилия	1
Имя	1
Отчество	1
Дата рождения	1
Место рождения	1
Пол	1
Дата регистрации	1

Image with superimposed blocks:

Дерево страниц