Singapore's War on Duplicate Images: What Officials, Experts and Key Figures Are Saying

Singapore's Info-communications Media Development Authority has escalated pressure on public-sector agencies and licensed platforms to implement automated duplicate image detection by the end of the third quarter of 2026, according to guidelines circulated to government departments earlier this year. The directive affects everything from property listing photographs on HDB's official resale portal to archival images held by the National Archives of Singapore on Canning Rise.

The timing is not incidental. Singapore's Smart Nation 2.0 framework, which the government formally advanced in late 2025, places data hygiene at the centre of its artificial intelligence ambitions. Duplicate or mismatched images—a listing showing the wrong flat, an archival record paired with an incorrect photograph—carry real consequences for a city-state where data accuracy underpins housing transactions worth hundreds of thousands of dollars each.

Why It Matters in a Tight Property Market

The HDB resale market gives this issue particular urgency. Median resale flat prices in mature estates like Toa Payoh and Queenstown crossed S$600,000 for five-room units in early 2026, according to data from HDB's quarterly flash estimates. Buyers relying on portal photographs to shortlist viewings have repeatedly flagged cases where images from previous listings remained attached to relisted units, creating confusion about renovation states and floor layouts.

The Council for Estate Agencies, which regulates property agents operating under the Estate Agents Act, has noted in its 2025 annual guidance to members that image accuracy in listings is a compliance matter, not merely a cosmetic one. Agents who allow duplicate or mismatched photographs to persist on platforms such as PropertyGuru or 99.co risk disciplinary referrals. The practical effect is that agencies are now investing in software that cross-references image hashes against existing databases before any listing goes live.

Beyond property, the National Library Board's digital collections programme, which manages more than two million digitised items through its NLB OverDrive and BookSG platforms, has identified duplicate image metadata as a cataloguing problem that inflates apparent collection sizes and complicates search results. NLB has been piloting perceptual hashing tools since January 2026 at its Lee Kong Chian Reference Library on Victoria Street, with a view to rolling out the system across regional libraries including Jurong Regional Library and Tampines Regional Library by year-end.

Experts Flag Deeper Technical Challenges

Computer vision researchers at Nanyang Technological University's School of Computer Science and Engineering have pointed out in published work that perceptual hashing handles near-duplicate images—those resized or lightly edited—far less reliably than exact duplicates. This matters because government photography archives frequently contain multiple crops or brightness-adjusted versions of the same original image, all of which carry different hash values under basic detection systems.

The more robust approach, according to peer-reviewed work out of NTU published in the IEEE Transactions on Multimedia journal in March 2026, involves training convolutional neural network models on locally sourced image datasets rather than relying on models built for Western or Chinese data environments, which may not generalise well to Singapore's specific mix of HDB architecture, multicultural signage, and tropical lighting conditions. The difference in detection accuracy in controlled tests was cited as exceeding 12 percentage points when locally trained models were used.

The Singapore Tourism Board, which curates photographic assets for Orchard Road, Marina Bay Sands, and Sentosa promotional campaigns, confirmed in a public procurement notice dated May 2026 that it is seeking a vendor to supply an enterprise-grade digital asset management system with built-in duplicate detection. The contract value was listed at up to S$2.1 million over three years.

For organisations navigating the transition, the IMDA's guidelines recommend a phased audit approach: first cataloguing all image libraries by source and date, then running hash-based screening, and finally applying machine-learning verification to the residual uncertain cases. Agencies that complete certification under the Trusted Media Framework by December 2026 will receive a streamlined data-sharing pathway with other public bodies—a meaningful incentive in a system where inter-agency data flows are tightly controlled. Private platforms have until March 2027 to comply with parallel standards under the Digital Platforms Regulation Bill currently before Parliament.