Singapore's Duplicate Image Crisis: The Key Decisions Ahead for Digital Archives and Property Records

Singapore's public sector is sitting on a growing problem buried inside its own servers. Across government databases — from HDB flat listing portals to the Urban Redevelopment Authority's property records and the National Archives of Singapore — duplicate image files have accumulated quietly over years of decentralised data management, creating storage bloat, retrieval errors, and in some cases, conflicting records that complicate decision-making in housing and urban planning.

The issue has sharpened in 2026 because Singapore's Smart Nation and Digital Government Office is mid-roll on its Whole-of-Government Data Strategy, a framework that commits agencies to interoperability and clean data pipelines by the end of this year. Duplicate images — photographs of flats, land parcels, infrastructure assets, and heritage buildings that exist in multiple versions across different systems — are now a material obstacle to that deadline.

Where the Problem Concentrates

The duplication is worst at the seams between agencies that digitised their own records independently before central data governance standards were formalised. The National Heritage Board's digitisation push for its collections at the former National Library on Victoria Street and the Asian Civilisations Museum on Empress Place created parallel image sets that were never reconciled with the National Archives' own holdings at Canning Rise. HDB's resale portal, which handles tens of thousands of flat images uploaded by sellers and valuers each year, has no automatic deduplication layer — meaning the same unit can appear in multiple listing cycles with the same photographs stored separately each time.

At the URA, planners working on development applications across precincts like Tengah and the Greater Southern Waterfront have flagged that site photographs submitted digitally by developers sometimes enter the system more than once through different application channels. The result is that reviewers occasionally pull outdated images when checking project status, introducing friction into an approval process that the government has spent years trying to accelerate.

The fix is not trivial. Perceptual hashing — the dominant technical method for identifying visually identical or near-identical images — works well for exact duplicates but struggles with photographs of the same location taken months apart, or images that have been cropped and resubmitted. Singapore's Government Technology Agency, GovTech, is currently evaluating two approaches: a rules-based deduplication engine that flags files for human review, and a machine-learning classifier trained on government image metadata. Neither is ready for full deployment.

The Decisions That Cannot Be Deferred

Three choices are coming to a head before the end of 2026. First, agencies must decide whether to centralise image storage under a single government cloud bucket — likely on the Government on Commercial Cloud platform already used by ministries — or allow agencies to keep local stores while syncing metadata to a master index. The centralisation route is faster to audit but requires agencies to give up administrative control they have held for years.

Second, there is a retention question. Government records management rules under the National Archives Act require that certain categories of image — particularly those tied to heritage buildings, compulsory acquisition proceedings, or planning appeals — be kept for defined periods, sometimes permanently. A deduplication process that deletes what it classifies as a duplicate could inadvertently destroy a legally required record if the classification is wrong. Legal counsel from the Attorney-General's Chambers will need to sign off on whatever deletion protocol is adopted.

Third, and most practically urgent for residents, is the HDB resale portal. Flat buyers in mature estates like Toa Payoh and Queenstown often rely on portal photographs to make preliminary assessments before physical viewings. If duplicate-cleaning goes wrong and deletes the only stored image for a listing, sellers bear a real inconvenience during what is already a high-stakes transaction in a market where four-room resale flats in central locations regularly transact above S$700,000.

GovTech has indicated — without specifying a timeline publicly — that a pilot deduplication exercise on a subset of HDB and URA records is planned before year-end. The pilot scope, the error tolerance built into the classifier, and how appeals will be handled if a legitimate image is incorrectly removed are the details that will determine whether the exercise builds or erodes public confidence in the government's digital estate management. The answers are expected before the Smart Nation Office's fourth-quarter progress review.