AI and Image Quality in Heritage Digitisation: Lessons for Family Archives

What the cultural heritage sector learned about image processing — and what it means for restoring your family photos.

On 14 September 2021, the ADOCHS project concluded with an online study day on Image and Data Processing in the Cultural Heritage Sector, organised jointly by CegeSoma (State Archives of Belgium), the KBR (Royal Library of Belgium), the Université libre de Bruxelles (ULB), and the Vrije Universiteit Brussel (VUB).

The event marked the end of four years of research into a question that sits at the intersection of archival science and artificial intelligence: how do you improve the quality of digitised heritage images at scale?

The answers developed for national institutions have since become the foundation of accessible AI tools that anyone can use to restore damaged family photographs.

The Problem: Digitisation Quality at Scale

When a national archive digitises tens of thousands of historical documents and photographs, quality control becomes a critical challenge. A scanner operator working through hundreds of items per day will inevitably produce inconsistent results — images that are too dark, too light, slightly blurred, or geometrically distorted.

For an institution like the KBR or CegeSoma, even a small percentage of low-quality scans represents thousands of unusable files. The ADOCHS project set out to develop automated methods to detect and correct these quality issues — methods grounded in computer vision and machine learning.

The research produced two categories of tools:

Detection tools — algorithms that automatically flag quality problems in digitised files: blur, low contrast, geometric distortion, noise, and physical damage such as tears or stains.

Correction tools — AI-powered methods to repair detected problems, either automatically or with minimal human intervention.

What the Research Found

The ADOCHS study day brought together specialists from institutions across Europe — including the BNF (Bibliothèque nationale de France), NIOD (Netherlands Institute for War Documentation), VIAA (Flemish Institute for Archiving), and the Ghent Centre for Digital Humanities — to present findings and discuss the future of AI in heritage digitisation.

Several conclusions emerged that extend well beyond institutional archiving:

AI Can Now Detect Damage That Human Reviewers Miss

Computer vision models trained on archival images can identify subtle quality problems — micro-blur, slight exposure inconsistencies, early-stage paper deterioration — that are invisible to a casual human review. For institutions processing large collections, this means automated quality control pipelines that are both faster and more accurate than manual inspection.

For individuals, this same capability is now embedded in consumer AI restoration tools. When you upload an old photograph to a restoration tool, the AI performs a damage assessment before applying any correction — identifying exactly what is wrong before deciding how to fix it.

Restoration and Digitisation Are Two Separate Steps

One key finding from the ADOCHS research: digitisation and restoration must be treated as distinct processes, with the original scan preserved separately from any corrected version.

This principle applies directly to personal archiving. When you digitise a damaged family photograph:

Save the raw scan as your archival master — unmodified, at full resolution
Create a working copy for restoration and sharing
Never overwrite the original with a restored version

The archival master is your insurance. Restoration techniques will continue to improve — the restored copy you make today may be superseded by a better tool in five years.

The Evolution of Organisations in the Age of AI

The afternoon session of the ADOCHS study day opened a broader discussion: as AI becomes capable of performing tasks previously requiring specialist human expertise, how do heritage institutions adapt?

The same question applies to families and individuals. AI has not eliminated the need for human judgment in archiving — it has shifted where that judgment is required. The decisions that matter are no longer technical (how to scan, how to correct colour balance) but curatorial:

Which photographs are worth preserving?
What context needs to be recorded alongside each image?
How should a family collection be organised so that future generations can navigate it?

These are questions that no algorithm answers. They require the knowledge that only family members hold.

From Institutional Research to Personal Practice

The tools that emerged from research programmes like ADOCHS — and from parallel work at institutions including the BNF, NIOD, and Beeld en Geluid — are now available to individuals through consumer AI applications.

The core capabilities developed for heritage digitisation translate directly into personal archiving:

Automated damage detection → AI restoration tools that identify scratches, tears, fading, and water damage before applying corrections

Batch processing pipelines → tools that can process an entire collection of scanned photographs in a single session

Quality scoring → automatic assessment of scan quality, with recommendations for rescanning where resolution or exposure is insufficient

Metadata extraction → AI-assisted identification of dates, locations, and in some cases faces — reducing the manual effort of describing a large collection

Building Your Family’s Digital Heritage

The ADOCHS project demonstrated that the gap between institutional archiving standards and personal practice is smaller than it appears. The methods differ in scale; the principles are identical.

A family collection treated with the same care as a national archive — digitised at archival resolution, restored from a preserved master, described with structured metadata, stored across multiple locations — will outlast any collection stored as an unsorted folder of smartphone photos.

The tools to do this are now accessible, affordable, and increasingly automated. What they require from you is a decision: to treat your family’s photographs as the irreplaceable historical record they are.

Heritage & AI