Singapore's Duplicate Image Problem: The Key Decisions Ahead for Government Databases and Property Listings

Singapore's push to consolidate its digital public infrastructure has surfaced a persistent, unglamorous problem: duplicate images clogging government databases, property listing platforms, and national identity systems. The issue is not new, but pressure to resolve it has sharpened as agencies accelerate their shift toward AI-assisted data processing and as storage costs climb alongside the Republic's ambitions to position itself as a regional technology hub.

The urgency is real. The Housing and Development Board maintains one of the largest visual archives in the public sector, cataloguing millions of flat photographs tied to resale listings, estate inspections, and renovation records. The Urban Redevelopment Authority similarly manages aerial and street-level imagery across planning zones from Woodlands to Tanjong Pagar. When duplicate images exist across these systems—the same photograph uploaded under multiple reference codes, or near-identical shots stored separately by different departments—the downstream effects include slower search retrieval, inflated cloud storage bills, and degraded accuracy in machine-learning models trained on that data.

Why the Next 12 Months Are Decisive

The Singapore government's Digital Government Blueprint has set targets for interoperability across agencies. GovTech, the agency responsible for driving that agenda, has been rolling out the Singpass-linked data infrastructure that underpins everything from CPF transactions to MyInfo property records. As that backbone matures, duplicate image data becomes more than a storage nuisance—it becomes a reliability risk. An AI model used to assess HDB flat condition for valuation purposes, for instance, performs worse when trained on datasets riddled with repeated frames.

The property market adds another layer of pressure. On platforms such as PropertyGuru and 99.co, which aggregate listings across HDB estates in Tampines, Bishan, and Buona Vista, duplicate images from sellers and agents are a known irritant. Listings for the same unit sometimes carry dozens of near-identical photographs, making automated quality-filtering—now standard on major platforms—a more computationally expensive task. With resale HDB flat prices having risen substantially over the past three years, the accuracy of listing data has direct financial consequences for buyers making decisions on transactions that regularly clear S$600,000 to S$900,000 for larger flat types.

Three decisions are converging right now. First, GovTech must determine which deduplication standard—perceptual hashing, deep-learning embeddings, or a hybrid—will become the interoperability baseline for agencies. Perceptual hashing is faster and cheaper; embedding-based methods catch near-duplicates more reliably but require heavier compute. Second, the Personal Data Protection Commission will need to clarify how image deduplication interacts with retention obligations under the PDPA, particularly for images tied to identity documents processed through immigration and the Immigration and Checkpoints Authority at Woodlands Checkpoint and Changi Airport. Third, private sector platforms will decide whether to align with any government standard voluntarily or wait for formal regulation.

What Agencies and Platforms Must Decide

The practical stakes are clearest in housing. HDB's resale portal processed tens of thousands of transactions in 2025, each generating documentation that includes photographic records. If the board adopts a centralised deduplication engine before the end of 2026, it would reduce storage overhead and improve the reliability of any computer-vision tools used for estate management—a goal the Smart Nation and Digital Economy Office has flagged as a priority area.

For private platforms, the calculus is partly competitive. A listing portal that cleans its image database faster builds better search and recommendation tools, which matters in a market where buyers comparing flats in Queenstown against those in Ang Mo Kio want clean, distinct visuals, not the same bathroom shot uploaded six times under different file names.

The decisions ahead are technical but the consequences are not. Agencies that defer will carry compounding storage costs and degrade the AI pipelines they are simultaneously being asked to build. Platforms that move early gain cleaner data and a compliance head start. GovTech is expected to publish updated data interoperability guidelines before the end of the third quarter of 2026—that document will likely be the first real signal of which direction the government intends to push, and whether the private sector will be expected to follow.