Step 1
Send the batch
Your server sends from 1 to 10 images in one request to the recognition method. Formats are JPEG and PNG; save the pages of PDF scans as images beforehand.
Automatic document classification
Identify the document type from a photo, with no pick-from-a-list step: NeuroVision names the country, type and form version, the side and page, tells a document from a selfie and sorts a batch of photos into verification steps. A document that does not fit gets a clear code, and the user gets a hint on what to retake.
Classification result Batch of 4 photos · one session
- Photo 1 detected Russian passport, 2006 version — main spread
- Photo 2 detected Russian passport — registration page
- Photo 3 detected Russian driving licence — front side
- Photo 4 detected Russian driving licence — back side
- Result sorted All photos are sorted into the steps of the scenario
- Classification is included in the “Document” package
- Up to 10 photos in one request
- 16,000+ document types from 195+ countries and territories
Data is stored in Russia Russian Software Registry No. 19972 Compliance with 152-FZ
Document type detection: what the service understands from a single photo
The user simply photographs the document
Classification is the first step of document recognition: the neural network looks at the whole image and, before extracting any fields, answers four questions. Automatic document type detection spares the user a “choose the document type” list and your team manual sorting. There is no need to specify the type and country in advance: the scenario only sets which documents are allowed at a step.
- 01
Which country
The issuing country code in three-letter ISO 3166-1 format — RUS, KAZ, UZB. It tells your system whether the person is a citizen or a foreigner before any field is read.
- 02
Which document and which form
Type and form version: a Russian passport of the 2006 or 2018 version, an international passport, an ID card, an old-style or new-style driving licence. The version determines where to look for each field on the form.
- 03
Which side and page
Front or back of a driving licence or ID card, the main passport spread or the registration page. If the customer photographed the wrong page, they find out straight away.
- 04
Document or selfie
The service assigns each image in the batch to a document or a selfie and sends it to its own scenario step. A photo with neither a document nor a face is rejected with a code.
- 16 000+ document types in the list
- 195+ countries and territories
- up to 10 photos in one request
Which documents the classifier tells apart
The classifier tells apart not only the type but also the form version, side and page. The full list by country is in the documentation.
-
Russian passports of all versions
All form versions from the list, including passports without a machine-readable zone, and individual pages: the main spread, registration and other spreads.
-
International passports and ID cards
Passports, electronic passports and national ID cards of CIS, European, Asian and Middle Eastern countries, front and back.
-
Driving licences
Russian driving licences from paper forms to plastic cards of the 2019 version, and foreign licences — with the front and back side detected.
-
Residence permits, temporary residence permits, visas and patents
Documents proving a foreigner’s right to live and work in Russia and other countries. Handy when a customer sends a passport and a permit together.
-
Official and special documents
Official and diplomatic passports, seafarer’s ID, refugee and stateless person documents, vehicle registration certificate.
-
Your document types
Need a document that is not on the list, or a company’s internal form? We will add the type from your samples. For questionnaires and applications we set up a field template.
Image batch: how the service sorts photos into steps and extracts fields
A customer sent a passport, a registration page and a driving licence in one go? Send them in a single request — the service determines the type of each photo and puts it on the right scenario step.
Step 2
Find the document in the photo
The service looks for a document in the frame. If the photo has neither a document nor a face, it is rejected at once and the user gets a hint.
Step 3
Detect the type, country and side
For each photo: the country, type and form version, the front or back side, the page number. The selfie is separated from the documents.
Step 4
Sort into scenario steps
You do not need to pass a step number: each photo goes to the step that expects exactly this document. A document the scenario does not expect is rejected.
Step 5
Extract fields — or skip
By default the service reads the document fields after classification. If you only need the type right now, turn off text recognition with a request parameter.
Step 6
Return the result
The response contains the outcome of each step: document type, country code, side, fields and error codes. The mode is synchronous or asynchronous with a webhook.
A separate scenario for each document: chosen by the first photo
-
One entry point for all documents
You pass the identifier of one scenario — the entry. At the first step the session accepts the documents of the first steps of all linked scenarios, and the widget shows the user which documents are accepted.
-
The front side picks the scenario
The first document that fits one of the scenarios sets the path: the session then follows its steps and checks. The back side, registration page and selfie do not pick a scenario.
-
The choice is final
The scenario will not change mid-verification. A document that fits none of the scenarios is rejected and the user retakes it.
-
The integrator sees both scenarios
API responses and webhooks carry the entry scenario and the selected scenario. Until a choice is made the second field is empty, so your server knows which stage the session is at.
A Russian passport needs checks against Russian databases, an international passport needs other steps. Linked scenarios work as one entry point, and the classification of the first document decides which path the session takes.
You can link scenarios whose first step is the front side of a passport or ID card and whose first steps do not overlap by type and country. How to set it up — in the linked scenarios guide.
What classification returns: fields in the API response
The classification result arrives in the same response as the recognised fields: your server gets the document type before anyone opens the photo. The fields are described in the session status documentation.
- Enable
Document step
outcome for each document photo
Classification
type—documentdocName— type and form version, for example “Passport (2006)”countryCode— country code, for exampleRUSisBackSide— back side or not
What else is in the step
- step status
- recognised fields with a confidence level
- forgery checks
- error codes
- number of attempts
- Enable
Selfie step
outcome for the face photo
Classification
type—selfiewithDoc— whether the selfie was taken holding the document
What else is in the step
- step status
- links to the images
- the result of matching the face with the photo in the document, as a separate step
faces, if the scenario includes it
- Enable
Session
outcome for the whole batch
Classification
results— steps in scenario orderschemaId— the entry scenarioselectedSchemaId— the selected scenario for linked ones
What else is in the step
- session status
- your client key
- processing time
Why a photo is not accepted: codes and hints for the user
-
[general] no documents on image
No document found in the frame. Hint: photograph the required document in full. -
[general] not a selfie
At the selfie step there is no face outside the document. Hint: keep the face in focus and clearly visible. -
[general] bad image
The file is not an image or the format is neither JPEG nor PNG.
-
[document] unknown document type
Document type not recognised. Hint: make sure you are photographing the required document. -
[document] bad document type
The document is recognised, but this step expects another — for example a driving licence instead of a passport.
-
[document] bad pages
The wrong page was photographed. Hint: photograph the main spread of the passport. -
[document] cropped
The document is not fully visible: the type and page cannot be determined from part of the form.
-
[fraud] multiple documents
Different documents of the same type in one session — for example two different passports. -
[fraud] manipulation detected
Different documents or documents of different people were uploaded in the session.
- Every classification rejection comes with a code and ready-made “what to do” text that you can show to the customer. Each step allows up to 6 free capture attempts.
- The hint texts come from the error code reference. How the service looks for signs of forgery is described on the authenticity check.
Where classification replaces manual document sorting
Anywhere a customer sends several documents at once or may send the wrong one.
-
Banks and MFIs
- A passport, a registration page and a second document arrive in one batch and fall into their steps by themselves.
- Example: Passport + registration + driving licence
-
Insurance
- For a policy and claim settlement the service separates a licence from a passport and the front from the back.
- Example: Passport + licence, both sides
-
Car sharing and rental
- The user photographs both sides of the licence without being asked “which side is this”, and the back costs nothing extra.
- Example: Licence: front and back
-
Employers of foreigners
- A foreign national’s passport is separated from a residence permit, temporary residence permit or patent, and the country is given by code.
- Example: Passport + patent
-
Hotels and tourism
- Guests from different countries show a passport or an ID card, and the scenario is chosen by the first document.
- Example: International passport or ID card
-
Marketplaces and platforms
- The document is checked at seller registration or payout: an extra or someone else’s document is rejected before any manual review.
- Example: Passport or ID card
Classification and 152‑FZ: collect only the documents you need
Only the documents you need
Classification helps avoid accepting more than needed
Under Parts 4 and 5 of Art. 5 of 152‑FZ, data must not be excessive in relation to its purpose. In the scenario you list the permitted documents — everything else is rejected and never reaches your system.
Retention period — yours
Set in the scenario
Under Part 7 of Art. 5, data is kept no longer than the purpose requires. The retention period of session data is set in the scenario, and you can delete a session earlier with an API method.
A file by type
Documents are sorted by type
If you work under 115‑FZ, identification data is kept for at least five years after the relationship with the customer ends. Documents sorted by type are easier to assemble into a customer file.
Processing in Russia
A cloud in Russia or your own perimeter
The NeuroVision cloud runs in data centres in Russia; with installation in your environment the photos never leave your infrastructure.
The data operator is your company
A guideline, not a legal opinion
The service helps you avoid accepting more than needed, but your company remains the personal data operator: you set the purposes, the list of documents and the retention periods. This is a guideline, not a legal opinion.
152‑FZ
a Russian developer, software in the registry, No. 19972
How much document classification costs
There is no separate line for classification: it is part of document recognition. You pay per KYC session with all scenario steps, with no subscription fee. Test access is 1 month.
Calculate the cost- 01
From 16 ₽ per session — the “Document” package: classification, recognition and document anti-fraud
- 02
10 ₽ — each additional document in the same session; backs of licences and ID cards — free
- 03
From 29 ₽ — “Document + liveness”: the same package plus a selfie, Liveness and matching the face with the document photo
How to embed classification: widget, REST API or your own server
Classification works with any integration method — no separate method or contract is needed.
- One request — up to 10 photos
- Type and country code in the response
- Rejection codes with a hint
- Test flow and documentation
Terms
Test access for 1 month, a personal manager, support under contract, pay only for checks.
Document classification glossary
- Document classification
- Determining from an image what document is in front of the service: country, type, form version, side, page. The first step of recognition — before field extraction.
- document classification
- Classifier
- A neural network trained to tell document types apart by the look of the form: the position of the photo, fields, coats of arms and the machine-readable zone.
- classifier
- Form version
- A version of a document of one type: for example, the Russian passport of the 2006 and 2018 versions. Different versions place the fields differently.
- template version
- Country code
- A three-letter state code under the international standard: RUS, KAZ, DEU. The same codes are used in the machine-readable zone of documents.
- ISO 3166-1 alpha-3
- Document code in the MRZ
- The first letter of the machine-readable zone under ICAO Doc 9303: P — passport, I, A or C — card-format ID, V — visa. More on the page MRZ reading.
- document code
- Scenario step
- A verification stage that expects a specific document or selfie. In the scenario, the permitted countries and document types are set for each step.
- scenario step
- Linked scenarios
- A group of scenarios with one entry point: the session picks a scenario by the user’s first document.
- linked scenarios
- Image batch
- Several photos in one request — up to 10. The service determines the type of each and sorts them into the steps of one session.
- batch
Send a batch of documents — we will show how it is sorted
Tell us which documents your customers send and in what form — we will check whether they are on the list and give you test access to the API and SDK for 1 month.
Can't put your request into words? Book a call and our team will contact you at a convenient time
Book a call
The company is included in the register of IT companies and the register of personal data operators of the Russian Ministry of Digital Development.
Address: Moscow, Presnenskaya Embankment 12 — Moscow City business centre, Federation Tower. Nearest metro stations: Delovoy Tsentr and Mezhdunarodnaya.
Read also
How document recognition, classification and authenticity checks work — in our blog
396 23
KYC Document Verification
AI-OCR in KYC: how automatic document recognition reduces manual review
434 21
KYC Document Verification
Why users retake documents: how to reduce the number of repeat attempts during a KYC check
439 10
AI OCR, IDP and Document Recognition
How to improve the accuracy of AI OCR and MRZ verification in KYC: metrics, errors and working solutions
456 13
KYC Document Verification
How KYC Systems Detect Forged Documents
Frequently asked questions about document classification
Answers about document types, photo batches, rejection codes and cost
It is the first recognition step: from an image the service determines the country, type and form version of the document, the side and page. Automatic document type detection is needed to read the fields correctly and to check that the document suits the verification step.
OCR reads text, while classification answers the question “what document is this”. Without it the service does not know where the surname is on the form and where the number is. In NeuroVision classification runs before field extraction in the same request.
No. The user photographs the document and the service determines the type itself. The scenario only sets which documents are allowed at a step; any other document is rejected with a code and a hint.
Yes, up to 10 images in one request to the REST API. You do not need to pass a step number: the service sorts the photos into scenario steps itself and separates the selfie from the documents.
JPEG and PNG images are accepted. Save PDF pages as images on your side. Scans and copies pass if the copy and printout checks are turned off in the scenario.
Yes. Text recognition is turned off with a request parameter — then the response contains the per-step result without extracted fields. This is handy for the initial sorting of an archive.
The step is not passed, the response returns a “wrong page” code, and the user sees a hint to photograph the main spread. Each step allows up to six free attempts.
Link the scenarios in your account. The session picks a scenario by the front side of the first document: for example, a Russian passport goes to checks against Russian databases and an international passport to its own steps.
More than 16 000 document types from 195+ countries and territories: passports, ID cards, driving licences, residence permits, temporary residence permits, visas, official and special documents. If the one you need is missing, we will add the type from your samples.
It is not charged separately: it is part of the “Document” package from 16 ₽ per session. An additional document in the same session is 10 ₽, and the first test month is free.
