HAVANK

Police

In HAVANK (The Automated Fingerprinting System Dutch Kollektie), finger and palm prints are converted to a code. The algorithm compares this code with the codes in the Criminal Justice Database, which contains the finger and handprints of suspects and convicts of crimes punishable by a prison sentence of four years or more.

Last change on 3rd of February 2025, at 15:51 (CET) | Publication Standard 1.0

• Publication category: High-Risk AI-system
• Impact assessment: DPIA
• Status: In use

General information

Theme

Public Order and Safety

Begin date

2022-07

Contact information

https://www.politie.nl

Responsible use

Goal and impact

The purpose of the algorithm is the automated comparison of fingerprints and handprints (hereafter fingerprints). Establishing a person's identity is of great importance in police processes. It is at the heart of investigations. Comparing many fingerprints is impossible and impracticable for a human being. Therefore, fingerprint comparison is automated. This algorithm contributes to efficient and effective policing, increasing the chance of being caught and thus safety.

The police use fingerprints to establish or verify a person's identity. This can include suspects, victims of crime, unknown deaths and witnesses. Forensics secures fingerprint traces at crime scenes and uses them to conduct investigations. The aim is to find the donor of the trace for truth-finding in criminal investigations and for evidence. HAVANK contains fingerprints of suspects and convicted persons of crimes punishable by four years or more in prison.

Considerations

Comparing the large quantities of fingerprints and traces manually is impossible and impracticable for a human being. Therefore, comparing fingerprints and traces is automated. This makes it much more likely to be recognised and less likely to make mistakes.

Human intervention

The algorithm fingerprints for identity determination produces a hit-no-hit result. This is done on the basis of a score: if the score is above a certain value then it is a hit, if the score is below a certain value then it is a no-hit.

The algorithm traces hand and fingerprints to produce a candidate list of scores. An expert examines the fingerprints presented by the system with the trace and investigates whether there might be a match. If there is no match then the investigation is closed with 'no recognition'. If there is a match then two other experts compare this match. They do this independently. Only when the three experts come to the same conclusion, formal individualisation occurs and is recorded in a report.

Risk management

- Only trained and authorised experts have access to the algorithm and fingerprints.

- To avoid bias, experts do not know which case they are working on, experts do not know from each other who is working on which case, and experts do not see each other's results.

- Automatic results are only generated above a reliable score threshold. A human checks if the data is correct or if there are any possible discrepancies.

- The results of the trace comparison algorithm are always assessed by at least three certified experts from the police according to a standardised work process under ISO 17025 accreditation (standard for testing and calibration laboratories). The outcome of the processing is thus the conclusion of three human experts.

Legal basis

The processing of the comparison of biometric data falls under the Police Data Act section 13(1c), identification of persons or things. In particular, the following articles are further explained in the context of processing:

- Article 5 (special categories of police data): The processing of police data revealing racial, ethnic origin, political opinions, religious or philosophical beliefs, or trade union membership, and the processing of genetic data, biometric data for the purpose of uniquely identifying a natural person, or data relating to health, sexual life and sexual orientation shall take place only if it is unavoidable for the purpose of the processing, in addition to the processing of other police data relating to the person and the data are adequately secured.

- Article 7a: A decision based solely on automated processing, including profiling, which produces adverse legal effects on or significantly affects the data subject shall be prohibited, unless provision is made for prior human intervention by or on behalf of the controller and for specific information to the data subject.

- Code of Criminal Procedure Articles 22(4), 27a, 29a, 55c, 61a and 190

- Penal Code articles 14c, 15a, 38, 38p, 38ab, 77f, 77ta, 77w, 77z and 447e

- Aliens Act 2000, sections 50 (identification) and 107 (processing fingerprints)

- Act and Decree on determining the identity of suspects and convicted persons

- Prüm Decision and other international treaties, for exchanging biometric data with foreign countries

- Code of Criminal Procedure article 55 c paragraph 4: Fingerprints may also be processed for the prevention, detection, prosecution and trial of criminal offences and establishing the identity of a corpse.

Impact assessment

Data Protection Impact Assessment (DPIA)

Operations

Data

Images of fingerprints and traces of hand and fingerprints.

Technical design

HAVANK converts an image of a fingerprint into an "arithmetic key" using an algorithm. This key is stored under a unique registration in a database. The image is also stored under the unique registration.

If a new fingerprint is presented to the system then the algorithm will again determine the "arithmetic key" and store it under a new unique registration. With this arithmetic key, a comparison is performed with all the arithmetic keys in the database. Using the algorithm, a score is then determined for each comparison.

The algorithm fingerprints for identity determination produces a hit-no-hit result. This is done on the basis of a score: if the score is above a certain value then it is a hit, if the score is below a certain value then it is a no-hit. The algorithm traces hand and fingerprints to produce a candidate list of scores.

The algorithm is not self-learning.

External provider

IDEMIA