Zurich Leads on Duplicate Image Removal — But Other Cities Are Closing the Gap

Zurich's city archive has quietly become a test case for how European municipalities handle a problem that barely existed five years ago: duplicate and AI-generated images contaminating official digital records. By June 2026, the Stadtarchiv Zürich had flagged and removed more than 14,000 duplicate or near-duplicate image files from its public-facing collections — a process that began in earnest in late 2024 after a routine audit revealed that roughly 8 percent of newly uploaded photographs shared near-identical pixel signatures with existing records.

The timing matters. Municipal archives across Europe and North America are under pressure after a wave of AI image generation tools made it trivially easy for individuals and institutions to upload synthetic or recycled visual content to government portals, planning applications and cultural heritage databases. The problem is no longer theoretical. In 2025, Amsterdam's Stadsarchief reported that a batch of 2,300 images submitted for a neighbourhood redevelopment consultation in the Jordaan district contained at least 190 duplicates or manipulated derivatives — some submitted by the same applicant under different registration numbers.

What Zurich Is Actually Doing

The Stadtarchiv, based on Alfred-Escher-Strasse, partnered with ETH Zurich's Data Analytics Lab in January 2025 to deploy a perceptual hashing system across its image management pipeline. Perceptual hashing — unlike standard checksums — can detect images that have been cropped, recoloured or slightly resized to evade basic duplicate filters. The collaboration cost the city CHF 340,000 over 18 months, according to budget documents published by the Stadtrat in March 2026.

The system is now embedded in the intake workflow for the Denkmalpflege, the city's heritage preservation office, which processes thousands of property images annually as part of Zurich's building inventory — a legal requirement under cantonal law for any structure older than 50 years. Since the ETH tool went live, the Denkmalpflege has cut manual image review time by an estimated 60 percent and rejected 1,100 submissions as duplicates or unacceptable derivatives in the first six months of 2026 alone.

Other Swiss cities have not moved as fast. Basel's cantonal archive is still using a manual spot-check process for duplicate detection, and Bern adopted a basic hash-matching tool only in April 2026. Neither has published removal statistics comparable to Zurich's.

How Zurich Compares Globally

The gap between Zurich and comparable mid-sized cities is real but narrowing. Seoul's metropolitan government launched a machine-learning-based image deduplication programme for its public planning portal in September 2025, covering approximately 2.4 million stored files. Vienna's Wienbibliothek im Rathaus began a pilot with the Austrian Institute of Technology in February 2026, though the project is limited to its digitised newspaper holdings for now.

London's situation is more fragmented. The 33 London boroughs each manage their own planning image databases, and there is no unified deduplication standard across them. A report by the Greater London Authority published in May 2026 found that Southwark and Tower Hamlets had duplicate image rates of 12 percent and 9 percent respectively in planning submissions received in 2025 — higher than anything Zurich has recorded since implementing its ETH-built tool.

The Zurich model carries costs that not every city can absorb. The CHF 340,000 price tag for the ETH partnership is substantial for smaller municipalities. Officials at the Stadtarchiv have indicated they are working on a licensing arrangement that could allow other Swiss cities to use the system at reduced cost from early 2027, though no formal agreement has been signed.

For residents and researchers using the Stadtarchiv's digital collections — accessible via the portal at stadtarchiv.ch — the practical effect is a cleaner, more trustworthy image library. Genealogists working in the Kreis 4 and Kreis 5 collections in particular had complained about encountering near-identical images catalogued under different dates and provenance tags, a problem the new system is designed to eliminate. The Stadtarchiv plans a full public-facing update to its search interface in September 2026 that will display image provenance scores directly in search results — a transparency step that, so far, no other European city archive has publicly committed to.