Detecting privacy conflicts on images

Water Board Hunze & Aa's


Water Board Hunze & Aa's


Anonimizer™ Visual Edition

The Dutch water board Hunze & Aa’s has an extensive library with photos of all kinds of places and objects in their management area: around 150,000 images of streams, lakes, waterways, locks, bridges, pumping stations etc. These images are paid for with public money and should therefore – according to the water board – be available to the public as open data. At the same time they have to deal with privacy legislation. If there are things on the photos that are directly traceable to the identity of people, you cannot just publish those photos. Now, you can choose to review all of those 150,000 photos one by one and filter out photos that do not comply with privacy legislation. That is a lot of work.

If the execution of a task requires human intelligence, but the handling of the task is inhumanly large, then the deployment of artificial intelligence is valuable. Assessing photos is a task that requires human intelligence. The number of 150,000 photos, however, makes it undesirable for a person to perform that task.

With our Anonimizer™ platform we solve the privacy problem for the water board. For this application, the Anonimizer agents have been trained to recognize people and cars in photos. This way a first selection of photos is made. With the help of computer vision and knowledge reasoning, the Anonimizer agents subsequently assess whether the faces of people or license plates on cars are recognizable. This way we filter out the photos that are not privacy proof and the Hunze and Aa’s water board is sure that they comply with the privacy legislation when publishing these images. Problem solved.