San Francisco Takes a Data-First Approach to Purging Duplicate Property Images From City Records — But Rivals Are Moving Faster

San Francisco's city government has a digital clutter problem. Hundreds of thousands of duplicate property images — many dating to the early 2000s digitization push — are clogging the databases maintained by the Office of the Assessor-Recorder on Van Ness Avenue, slowing permit processing times and creating discrepancies in public-facing property records that affect everything from Muni right-of-way assessments to affordable housing audits in the Tenderloin.

The city's Department of Technology confirmed this spring that a formal duplicate-image-replacement initiative is now underway, folded into a broader data modernization program that began in earnest in early 2026. The effort is long overdue. Property record systems in dense urban cores accumulate image redundancy at scale — a single parcel in the Mission District can carry five or six near-identical facade photographs taken by different city inspectors over a decade, each stored separately and cross-referenced inconsistently.

Why This Moment Matters

The timing is not accidental. San Francisco's housing production emergency — the city fell well short of its state-mandated Regional Housing Needs Allocation targets for the 2023–2031 cycle — has put intense pressure on permit offices to speed up approvals. Duplicate or mismatched property images are a documented source of delay: staff at the San Francisco Planning Department on Stevenson Street have reported spending significant administrative time reconciling image records before issuing discretionary review decisions. Clean, deduplicated visual records shorten that reconciliation step.

The technology driving the cleanup is a perceptual hashing algorithm layered on top of the city's existing Accela permitting platform, the same system used by more than 500 jurisdictions across North America. The algorithm flags images with more than 95 percent pixel similarity and queues them for human review before deletion or replacement. The city's IT team, working out of the Department of Technology offices on Folsom Street, began a pilot run covering Assessor District 7 — which includes West Portal, Forest Hill and St. Francis Wood — in March 2026. Early results from that pilot have not yet been made public.

The San Francisco Public Library's Government Information Center on Larkin Street has separately digitized neighborhood planning documents going back to 1967, and archivists there have used open-source deduplication tools for years. That institutional experience is now being tapped by the Department of Technology as it scales the property-records project citywide.

How San Francisco Compares Globally

Other cities started earlier and finished faster. Amsterdam's municipal land registry, the Basisregistratie Adressen en Gebouwen, completed a full deduplication sweep of its property image database in 2023 after a 14-month project. Seoul's Smart City Division wrapped a similar exercise covering approximately 4.1 million cadastral records in late 2024, using computer vision tools built in-house by the city's digital government bureau. Both cities now maintain real-time deduplication pipelines that prevent redundant images from entering the system in the first place — a prevention-first model San Francisco has not yet adopted.

London's Ordnance Survey partnership with borough councils represents a third model: centralized image standards enforced at point of capture, meaning field inspectors upload photographs through a single app that rejects near-duplicate submissions automatically before they reach any database. San Francisco's inspectors still use a patchwork of mobile tools with no unified upload standard, according to the Department of Technology's own 2025 annual report on digital infrastructure.

The gap matters financially as well as operationally. Storage costs for redundant municipal image data are not trivial. Across comparable U.S. cities, industry analysts have estimated that unmanaged image duplication can inflate cloud storage expenses by 15 to 30 percent annually — figures that translate to real budget line items at a time when San Francisco faces a projected general fund shortfall heading into fiscal year 2027.

The citywide rollout, if the Assessor District 7 pilot proves successful, is expected to reach all 11 assessor districts by the end of calendar year 2026. Residents and property owners who want to flag discrepancies in their own parcel's image records can file a correction request through the Assessor-Recorder's online portal, where the turnaround time for reviewed submissions is currently listed as 30 business days.