Restoring Old Photos to Improve Face Recognition

Patricia had 847 photographs in her family archive. She knew approximately 200 of the subjects. The other 647 photographs showed people she couldn't identify — relatives from earlier generations, the network of friends and extended family that constituted a life, captured in photographs that had become separated from their context.

She came to me with an unusual request: she wanted the photographs restored specifically to help with identification, not primarily for display or preservation. She was running the restored images through genealogical databases and facial comparison tools.

Face Recognition Requirements vs. Display Requirements

Optimizing a photograph for face recognition is different from optimizing for display. The differences matter for restoration approach.

For display: The overall aesthetic of the image matters. Tone, color, grain — the whole frame contributes to how the photograph looks and feels.

For face recognition: The face region alone matters, and it needs to meet specific criteria: sufficient resolution (at least 128×128 pixels of face area, ideally more), consistent lighting without extreme contrast, frontal or near-frontal orientation.

AI face restoration models (GFPGAN, CodeFormer) produce results that often perform better in face recognition than the pre-restoration originals, even when the restoration is not perfect. This is because the face models bring facial geometry and feature relationships into the expected range for comparison algorithms.

Scanning for Identification Purposes

When restoration is aimed at identification, scan resolution matters even more than usual. A face that's 100×100 pixels in a standard scan becomes 200×200 in a 2× higher-resolution scan — and the difference in face recognition accuracy can be significant.

For small prints containing faces you want to identify: scan at 1200-2400 DPI. The file sizes are large but the identification results are better.

Genealogical Face Comparison Tools

Several services offer face comparison for genealogical purposes:

MyHeritage includes face recognition for family tree photos.

FindMyPast offers photo comparison features.

Google Photos (not genealogy-specific) has strong face grouping that can help identify who appears together across a collection.

All of these work better with restored photographs than with degraded originals.

Start the restoration process that enables identification at our photo restoration tool.