Singapore's AI-Powered Push to Purge Duplicate Images From Government and Public Platforms Picks Up Speed This Week

Singapore's Infocomm Media Development Authority confirmed this week that a multi-agency pilot to detect and remove duplicate images across public-facing digital platforms has entered its second phase, bringing automated deduplication tools to at least four government-linked repositories by the end of July 2026. The move follows months of quiet testing and signals a broader reckoning with how fast-growing digital archives are ballooning with redundant visual content.

The timing is deliberate. Singapore has accelerated its Smart Nation 2.0 agenda over the past eighteen months, and bloated image databases — many containing thousands of near-identical or outright duplicated files — have become a measurable drag on storage costs, search performance, and user experience. For a city-state positioning itself as a regional AI hub, tolerating avoidable digital clutter is increasingly seen as a credibility issue.

Where the Problem Shows Up

The most visible friction point has been property listings. On HDB's official resale flat portal, buyers browsing units in Toa Payoh, Tampines, and Queenstown have long encountered listing galleries where agents upload the same photograph multiple times — occasionally dozens — inflating page load times and obscuring genuinely distinct images of a flat. A review conducted earlier this year found that a meaningful share of image assets across listing-adjacent public platforms contained near-duplicate files, though the IMDA has not released a precise figure for the government estate-linked inventory specifically.

The National Library Board's digital collection, accessible through its NLB OverDrive and BiblioAsia platforms, faces a parallel challenge in its historical photograph archive. Digitisation drives over the past decade have pulled in images from multiple donor sources, and the same photograph — say, a 1960s street scene along Orchard Road or a kampung in Bukit Timah — sometimes exists in three or four slightly different scan resolutions, each treated as a separate file. Librarians and archivists have flagged the issue internally for years.

The IMDA pilot draws on perceptual hashing and convolutional neural network matching, technology that compares images not by filename or metadata but by visual content. Similar systems have been deployed by the National Archives of Singapore for a subset of its photographic collection since late 2024, giving the current rollout a precedent to build on.

What the Week's Developments Mean in Practice

On Tuesday, GovTech published a brief technical note — part of its regular developer documentation updates — outlining application programming interface endpoints that agencies can use to flag images for deduplication review before they are ingested into centralised storage. The note specified a threshold sensitivity setting, allowing agencies to distinguish between exact duplicates and near-duplicates, which may have legitimate archival reasons for coexisting.

The practical stakes are not trivial. Cloud storage costs for the Singapore government's whole-of-government commercial cloud contracts, awarded to providers including Amazon Web Services and Google Cloud under the Government Commercial Cloud framework, run into tens of millions of dollars annually. Removing redundant files does not eliminate those contracts, but it does reduce the volume of assets requiring backup, indexing, and retrieval bandwidth — costs that compound at scale.

For private-sector operators adjacent to government platforms, particularly property agencies submitting listings through the HDB InfoWEB system and the Urban Redevelopment Authority's REALIS portal, the new protocols carry a practical implication: images submitted with detectable duplicates within a single listing batch may be flagged and returned for resubmission from August 1, according to the GovTech documentation.

Agents and small agencies operating out of offices along Tanjong Pagar and the Beach Road cluster should audit their standard listing photography workflows before that deadline. Most professional photography tools — including Adobe Lightroom and several property-specific CMS platforms already in common use here — carry built-in duplicate detection that, if switched on, would catch the majority of cases before upload. The technical fix, in most cases, is simpler than the awareness gap that has allowed the problem to persist.