Appearance
Quality control of digitised images means systematically verifying that every file meets your project's technical and descriptive targets before it becomes the master copy of an irreplaceable object. The most effective approach combines automated checks that run across 100% of a batch (completeness, dimensions, bit depth, filename rules, embedded metadata) with a sampled human review of 5-10% plus every automatically flagged file. Done well, QC is not a one-off glance at a thumbnail; it is a documented, repeatable process you can defend to an auditor years later.
What exactly are you checking for?
QC splits into four families of defect, and a good checklist names all four explicitly:
- Completeness and identity — every expected page or object present, in order, with a filename matching your naming convention and no duplicates or skipped sequences.
- Image quality — focus/sharpness, even illumination, no clipping in highlights or shadows, no moiré or compression artefacts, correct orientation.
- Colour and tone — neutral greys reading as neutral, tonal range matching a captured target, white balance consistent across the batch.
- Technical and descriptive metadata — correct resolution, bit depth, colour profile embedded, and capture metadata present and accurate.
How much of a batch should I actually inspect?
Inspect everything cheaply, then inspect a sample expensively. Automated rules are fast enough to run on every file, so completeness, dimensions and metadata should be 100% verified by script. Human eyes are the scarce resource, so reserve them for a stratified sample — typically 5-10% per batch — plus 100% of anything a rule flagged.
A quick ImageMagick pass surfaces the cheap defects across a whole folder:
bash
# Report geometry, colourspace and bit depth for every TIFF
for f in *.tif; do
identify -format "%f %wx%h %[colorspace] %z-bit\n" "$f"
done > qc_dimensions.txt
# Flag any file that is not the expected 600 ppi master size band
identify -format "%f %x\n" *.tif | awk '$2 < 590 || $2 > 610 {print "RES?", $0}'Which tools verify quality objectively?
| Check | Tool | What it gives you |
|---|---|---|
| Dimensions, bit depth, colourspace | ImageMagick identify | Fast batch geometry report |
| Embedded metadata | ExifTool | Verify/inject EXIF, IPTC, XMP fields |
| Resolution (SFR/MTF) | Imatest or GoldenThread + target | MTF50, sampling efficiency vs FADGI |
| Colour/tone accuracy | DICE/Universal Target | Delta-E and tonal deviation |
| Visual artefacts | Calibrated monitor + viewer | Human review of the sample |
ExifTool is the workhorse for the metadata family:
bash
# Confirm every master embeds an ICC profile and the right resolution
exiftool -ColorSpace -ProfileDescription -XResolution -r -ext tif .How do I catch colour casts and uneven lighting?
Shoot a calibrated target — a Universal Test Target or X-Rite chart — at the start of every session and whenever you change the rig. Then measure the neutral patches. If the R, G and B values for a grey patch drift apart by more than a few points, you have a colour cast; if they drift together as you move along the greyscale, your tone curve is off. Uneven lighting shows up as the same neutral patch reading differently when moved to each corner of the copy field, so test the corners deliberately.
How do I document QC defensibly?
Treat the QC log as part of the archival record, not a sticky note. For each image or batch record at minimum: identifier, date, operator, each check performed, pass/fail, and the remediation taken. A flat CSV is enough to start and travels with the files:
csv
filename,date,operator,completeness,focus,colour_target,metadata,result,notes
ms_042_001.tif,2025-02-18,ereed,pass,pass,pass,pass,pass,
ms_042_017.tif,2025-02-18,ereed,pass,fail,pass,pass,reshoot,soft top-rightStoring this alongside the images means anyone can later reconstruct exactly why a file passed.
What do I do with a failure?
Decide remediation rules in advance so QC does not stall. Soft focus, wrong framing or a missing page means re-shoot while the object is still on the cradle — cheap now, expensive once it is reshelved. A metadata error means re-run the injection script. A colour issue traceable to the session means reprocess the RAW with the corrected target. Record the action, then re-QC the corrected file so the log shows a clean pass.
Key Takeaways
- Run automated checks on 100% of files and reserve human review for a 5-10% sample plus all flagged files.
- Cover four defect families: completeness, image quality, colour/tone, and metadata.
- Capture a calibrated target every session so colour and resolution claims are measurable, not subjective.
- Use ImageMagick and ExifTool for fast batch verification of geometry and metadata.
- Set project-specific thresholds from your own best images when a published number does not fit.
- Log every check, operator and result alongside the images to keep QC defensible.
- Define remediation rules in advance and re-QC anything you correct.
Frequently Asked Questions
What percentage of a batch should I QC?
Inspect 100% of images for completeness and filename correctness using automated checks, then visually review a representative sample of 5-10% of each batch, plus every image flagged by an automated rule.
Which open-source tools verify image quality objectively?
ImageMagick and ExifTool handle dimensions, bit depth and metadata; the DICE OpenDICE target plus a tool like Imatest or the free GoldenThread alternative measure resolution, tone and colour accuracy against FADGI metrics.
How do I detect colour casts in a digitised batch?
Photograph a calibrated colour target in the same session, measure the neutral patches, and confirm the R, G and B values stay within a few points of each other; a consistent offset across patches indicates a white-balance cast.
Should QC happen during capture or after the whole batch?
Both. Run quick in-session checks (focus, framing, target) so you can re-shoot immediately, and run a deeper automated and sampled review after capture before files leave the imaging station.
How do I document QC so it is defensible later?
Log every check, who ran it, the date, the pass/fail result and any remediation in a structured file (CSV or a database), and keep it alongside the images so an auditor can reconstruct your decisions.
What is an acceptable focus or sharpness threshold?
There is no universal number, but FADGI 3-star and 4-star tiers specify SFR/MTF50 limits; in practice, set a project baseline from your best images and reject anything measurably softer than that baseline.