A small image-processing project at a German online retailer offers a practical warning for quality teams: when an image carries mandatory text, a clean-looking result is not evidence that the information survived. Justin Flick of SOMUVA UG, which runs the German-language shop vapechamp.de, described the case in an article on text fidelity in processed images. The company wanted to turn phone photos of carton back panels into straightened images on a white background for product pages. The panels carry legally required manufacturer and importer details, warning texts and contact information. The aim was a better appearance without changing the printed information.
The team defined the measurand as the printed text, not the appearance of the image. Its first attempt used a generative image model, with an explicit instruction to keep every printed character, number and symbol identical. The output looked professional: sharp print, even light and a clean background. A comparison with the source found otherwise. The digits 2801 had become 2001; a phone number contained wrong digits; a comma had turned into a full stop, and elsewhere a full stop had become a comma. Because the changes were plausible, they were hard to spot at a glance. For mandatory information, that was not a cosmetic issue. The generated images were discarded and none was published.
The lesson was that a generative model produces text that looks plausible rather than reproducing what was printed. Visual inspection is a weak check against plausible errors, which is why it cannot serve as the reference. That distinction matters beyond one retailer. In image workflows that support regulated labels, safety warnings or contact details, the acceptance criterion has to be the characters, not the polish of the image.
A deterministic chain from photographed pixels
The replacement was a deterministic chain. The carton face is rectified with a perspective transform from 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 gentle and edge-preserving. Text blocks are marked as protected regions in which pixels pass through the background clean-up unchanged. The result is placed on a white background with a soft shadow. Nothing on the carton is redrawn or re-typeset; every character in the final image is derived from the photographed pixels.
Two gates against an early reference
For each image, the team kept a reference: the same crop after rectification and the illumination and white-balance step only, before denoising or clean-up. Every processed carton face was compared with that reference before it was placed on the background. Two measurement principles were used. The first was an OCR diff. Both images were read by an open-source OCR engine with German and English language data after upscaling, and the word sequences were aligned so every deviation could be 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 such 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 showed that classical filters are not automatically safe. A speckle filter had turned 130 into 30 and removed colons. A threshold clean-up step had erased the dot in a web address. Both stages were dropped. The team's formulation was direct: any stage that defines what counts as noise can also classify a full stop as noise. That point matters for quality teams that treat AI tools as the main risk and conventional filters as benign. The case put both through the same reference check.
OCR has its own measurement uncertainty
The OCR diff was itself a measurement system with its own error. On small print, the OCR misread the reference image as well, for example reading a 5 as an S. A raw OCR diff therefore reported disagreements that the processing had not caused. Taken at face value, those reports would lead to false rejects; ignored wholesale, they would hide real changes and produce false accepts. Flick compared the problem with a gauge study: before attributing variation to the part, you separate out the variation of the measurement system. The OCR's misreads of the reference had to be separated from changes introduced by the processing. A second, independent principle, the ink check, and a separate review step did that work.
The approach matches Tsipursky's third step, to identify the source of the disagreement rather than assuming the person or the model must be wrong. The operating principle was to treat each disagreement as a stop condition to resolve, not as a similarity score to average away. That changes the review question from whether the output looks close enough to whether every deviation has an explained cause.
Second review, backups and the published 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 files as actually delivered were downloaded and compared with the team's 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 deserves the same check as every step before it.
Practical checks for quality teams
The case yields a set of practical checks for quality teams:
- Define the measurand. For images that carry mandatory text, the characters matter, not the appearance.
- Keep a separate reference from the earliest reliable processing step, and never compare an output only with itself.
- Use two measurement principles, such as an OCR text diff and a pixel-level ink check.
- Characterise the OCR's own errors on the reference before trusting its verdicts.
- Treat any disagreement as a stop, not a score.
- Test classical filters as rigorously as AI tools.
- Add an independent second review, and remove what cannot be verified.
- Back up before every change and pixel-compare the published files.
- Improve the input: frontal, unwrapped, frame-filling photos in daylight make later steps easier.
Generative models remain useful for images where nothing has to be read. For images that carry mandatory information, the team now uses classical processing plus a text comparison against the original. Tsipursky posed a question quality leaders should be able to answer: when the machine and a qualified person disagree, do we know who was right, why, and what we 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. The SOMUVA case is small and outside the factory, but it shows how far the principle travels: an automated result is only as trustworthy as the independent reference it has been checked against.
