When an image is cleaned up for an online product page, the visible result can look more professional while the information it carries becomes less reliable. A case described by Justin Flick of SOMUVA UG, which runs the German-language shop vapechamp.de, makes that risk concrete. The company needed to turn phone photographs of carton back panels into straightened images on a white background. Those panels carry legally required manufacturer and importer names and addresses, warning texts, and contact details such as phone numbers. The images could be improved, but the printed information had to remain unchanged. In measurement terms, Flick writes, the measurand was the printed text, not the appearance of the image.
The case echoes an argument by Gleb Tsipursky in Metrology News: quality leaders should measure disagreements between automated inspection and qualified human judgment before reducing human review. Although the carton project sits outside a factory, the same principle applied. An automated result is only as trustworthy as the independent reference against which it has been checked.
A plausible result that failed the text test
The first attempt used a generative image model. The team supplied a straightened carton image and explicitly instructed the model to keep every printed character, number and symbol identical. The output looked polished, with sharp print, even lighting and a clean background. Comparison with the source revealed changes: the digits “2801” had become “2001,” a phone number contained wrong digits, and commas and full stops had been exchanged in places. These were not glaring errors. They were plausible substitutions, which made visual inspection a weak safeguard.
Because the altered text was mandatory information, the team discarded the generated images and published none of them. Flick's conclusion is that a generative model produces what text plausibly looks like, not necessarily what was printed. A visual check cannot serve as the reference when the failure mode is a believable error.
A deterministic chain that keeps photographed pixels
The replacement was a deterministic processing chain. Each carton face is rectified with a perspective transform based on four corner points. Uneven lighting is flattened by estimating the background illumination and dividing it out, followed by white balancing to a target tone. Denoising is kept gentle and edge-preserving. Text blocks are marked as protected regions, so their pixels pass through background clean-up unchanged. The result is then placed on a white background with a soft shadow. Nothing on the carton is redrawn or re-typeset. Every character in the final printed panel is derived from the photographed pixels.
An early reference and two independent gates
For each image, the team kept a reference: the same crop after rectification and the illumination and white-balance steps, but before denoising or clean-up. Each processed carton face was compared with that reference before being placed on the background. The comparison used two measurement principles. The first was an OCR diff. Both images were upscaled and read by an open-source OCR engine with German and English language data; the resulting word sequences were aligned and every deviation listed. The second was an ink check that did not depend on reading text. Pixels clearly darker than the carton background in the reference counted as print. If those pixels became light in the processed image and most of a connected group disappeared, the group was flagged as a lost character.
The gate also exposed failures in classical processing. A speckle filter turned “130” into “30” and removed colons. A threshold clean-up step erased the dot in a web address. Both stages were dropped. The lesson is that any step deciding what counts as noise can also decide that a full stop is noise. Classical filters therefore needed the same rigorous testing as AI tools.
OCR is a measurement system with its own error
An OCR diff is not a perfect judge. It is itself a measurement system with its own uncertainty. On small print, the OCR misread the reference image, for example reading a “5” as an “S.” A raw OCR diff therefore reports disagreements that the processing did not cause. Taking those at face value would produce false rejects; ignoring them wholesale would hide real changes and produce false accepts.
The team used a gauge-study analogy. Before attributing variation to the part, one separates the variation of the measurement system. In this case, the OCR's misreads of the reference had to be separated from changes introduced by processing. A second, independent principle—the ink check—and a separate review step did that work. This follows Tsipursky's third step, to identify the source of a disagreement rather than assuming that the person or the model must be wrong. Each disagreement was treated as a stop to be resolved, not as a similarity score to be averaged away.
Second review, backups and the delivered file
A separate automated review, independent of the processing chain, checked the images in two rounds. In the first round, a composite image failed because one inserted line could not be verified reliably. The team removed the composite and used the genuine information block from the carton instead. Before any image was replaced, the previous state was backed up so it could be restored. After publication, the team downloaded the files as delivered and compared them with its own copies. At reduced resolution, the mean absolute greyscale difference was well under one grey level. The published file is the last link in the chain and was checked like every step before it.
What quality teams can take away
The practical lessons begin with defining the measurand. For images that carry mandatory text, the characters matter, not the appearance. A separate reference image should come from the earliest reliable processing step, and an output should never be compared only with itself. Two measurement principles—such as an OCR text diff and a pixel-level ink check—provide a stronger basis for judgment. The OCR's own errors on the reference must be characterised before its verdicts are trusted. Any disagreement should be treated as a stop, not a score.
Classical filters should be tested as rigorously as AI tools. An independent second review should be added, and anything that cannot be verified should be removed. Backups should precede every change, and published files should be pixel-compared with the team's own copies. Better input also helps: frontal, unwrapped, frame-filling photographs taken in daylight make later steps easier. Generative models remain useful for images where nothing has to be read. For images carrying mandatory information, the team now uses classical processing plus a text comparison against the original.
Tsipursky's closing question for quality leaders is whether, when a machine and a qualified person disagree, they know who was right, why, and what was changed afterward. For processed images, the question is the same. When the output and the reference disagree, the answer should be established before anything is published.
