The procedure, in the order that decides the result
A colour comparison is a procedure with four steps, and skipping the second one is why most of them end in an argument rather than a decision. The aim is not to capture the true colour of an item, which no warehouse camera can do, but to capture two frames that differ only in the thing you are testing.
- Photograph the item with a known reference object in the same frame, and use the same reference every time.
- Take a second frame under a different light without moving the item or the camera position.
- Compare the item against the reference inside each frame, never against your memory of the previous frame.
- Decide only on the difference that survives that comparison in both directions.
The second step costs nothing and is almost always skipped. Two frames taken minutes apart from the same position, one under warehouse lighting and one near a window, are a controlled pair. Two frames taken from different angles on different days are two unrelated pictures, and any difference between them could belong to the camera, the light, the angle or the item, with no way to tell which.
The procedure is short because the useful part of it is a refusal. Three of the four steps exist to eliminate explanations other than the item itself, and a comparison that skips them produces an impression that feels like evidence and is not.
Two frames are the minimum and three are usually better, because a comparison made of two points cannot show whether a difference is stable. The third frame, taken under the first light again a few minutes later, tells you whether the camera drifted between shots, and that drift is the failure which makes two-frame comparisons disagree with themselves when they are repeated the next day.
What to put in the frame beside the item
The reference object is the sample, and a frame without one has no scale for colour at all, because every camera decides what white is at the moment the shutter closes and keeps that decision to itself.
| Reference | Calibrates | Cannot calibrate |
|---|---|---|
| Plain white paper | Which way white was pushed | Hue under coloured light |
| A printed colour chart | Hue and saturation across patches | Patches it does not contain |
| Your own hand | What skin should look like | Anything about the item |
| A second copy of the item | Batch-to-batch variation | Whether either matches the listing |
The last row answers the question buyers actually ask, which is whether the item is wrong or whether the batch simply differs. Two units of the same order photographed together separate those two cases inside a single frame, and that frame is worth more than a dozen solo shots.
A reference also has to be something the warehouse already has. A printed colour chart may not be within reach, while a sheet of paper, a coin and a second unit of the same order usually are, and a request that depends on equipment nobody owns tends to come back unanswered rather than refused. Naming the reference you want is part of writing the request, not an afterthought added once the reply disappoints. Ask for one reference rather than three, so that the frame stays readable at a glance.
Where the error gets in
Four sources of error sit between the item and your screen, and not one of them publishes a magnitude. Camera white balance is the largest and the most automatic: the camera solves for white on its own, so the same coat photographs differently against a wall and against a carton. Reading the other three sources before blaming the item is what stops a claim that cannot be supported.
- Automatic white balance: a single decision made by the camera and recorded nowhere in the file.
- Mixed sources: warehouse lighting and daylight entering the same frame, each with its own colour.
- Compression: strongly saturated reds and blues lose detail first, which is where most colour complaints live.
- Your screen: an uncalibrated display, a night schedule, or a brightness setting changed since the last judgement.
How far each of those four sources moves a colour is not published by anyone. This page prints no tolerance figure; it describes the order to compare in rather than a threshold to compare against.
One consequence follows immediately. A colour difference that exists only on your screen is not evidence of anything, and a claim built on it will be answered with a request for another photograph, which will produce another screen and another opinion.
The order of those four matters when you are trying to explain a disagreement. White balance and mixed sources move the whole frame together, while compression and display settings move saturated areas and dark areas by different amounts. A uniform shift therefore points at the light, and a patchy one points at the file or at the screen you are reading it on, which is a different conversation from the one about the item.
What a colour comparison can support
Two outcomes are usable and the rest are not. If the item differs from the reference in the same direction in both frames, the difference is in the item. If the difference appears in one frame and vanishes in the other, the difference was in the light, and there is nothing to raise.
- Supportable: two copies of the same item differ from each other inside one frame.
- Supportable: the item contradicts a written colour name in the listing, in both frames.
- Unsupportable: two photographs of one item taken on different days, from different angles, under different light.
Comparing raw pixel values across two frames compares two camera decisions. Comparing each frame against the reference inside it compares the item, because both frames get corrected by the same operation.
The arithmetic that makes this work is a ratio rather than a reading. For each frame, divide the item channel values by the reference channel values on the same channel, using the 0 to 255 scale the file already uses. A hue shift that survives the division is in the object, and a shift that disappears was in the lighting. Run the division on both frames, because a conclusion drawn from one frame is a reading, and a reading is exactly what the reference was meant to replace.
The cases where you should not judge colour at all
Some materials cannot be judged from a still image at any level of care. Anything iridescent, brushed or heathered shifts hue with the angle of the light, prints and gradients are judged on a fraction of their surface, and a garment described as washed or vintage has a colour that is not supposed to be uniform in the first place.
In those cases the useful request is not a better photograph but a different question entirely: a composition label, a measured length, or a statement about the variant code the warehouse received. A colour argument about a heathered fabric will spend the storage window and settle nothing, which is the worst combination available.
Where colour genuinely is the question, the [warehouse QC pages](/table/qc/) treat it as the weakest photographable property, and that ranking is the one to keep in mind. The [QC stage walkthrough](/steps/qc-photos/) describes when the window for asking closes, and colour is not a reason to reopen it.
Print is the clearest example of the boundary. A garment whose pattern runs across a seam cannot be judged from a frame that shows part of the pattern, and a gradient printed across a large panel is judged on whichever fraction of it the camera happened to fill. Where a print is the concern, the usable questions are about the pattern matching at the seams and about the position of the print on the panel, both of which are geometry rather than colour.
Worked example: normalise each frame by the reference inside it, dividing the item channel values by the reference channel values on the same channel across the 0 to 255 range. A hue shift that survives the division is in the object; a shift that disappears under it was in the lighting, because both frames were corrected by the same operation. (Worked example)
This note explains the mechanics; the list is where the items are.