You measured the cabinet face at 600 mm. Your buyer scales it off your photo and gets 618. Neither of you is lying — the lens is. Knowing how to photograph a product for measurement is the difference between a spec image that survives an argument and one that starts one.
Two separate things bend the numbers in a product photo, and most people only know about one of them. Lens distortion is the optical error that makes a straight edge image as a curve. Perspective error is the geometric error that makes closer parts of the object image larger. They have different causes, different fixes, and only one of them can be corrected in software after the fact.
The Two Error Families
| Lens (geometric) distortion | Perspective error | |
|---|---|---|
| What it does | Straight edges bow outward (barrel) or inward (pincushion) | Nearer parts image larger; parallel edges converge |
| Worst case | Ultra-wide and zoom extremes; phone 0.5× modes | Camera close to the object, or tilted off-axis |
| Fixable afterwards? | Yes — lens profile correction, if the profile exists | Only partly; keystone correction re-warps and softens |
| Standardised measurement | Yes — ISO 17850 dot-chart protocol | No standard; it is pure geometry |
| Telltale sign | The horizon or a shelf edge curves | The far end of the product looks narrower |
ISO 17850:2015 specifies the protocol for measuring a digital camera's geometric distortion using a dot chart, and it applies to camera phones as well as cameras. Imaging labs report distortion against that protocol, which is why lens-profile corrections exist at all — someone measured the lens first.
Distance Matters More Than Anything Else
Here is the number that decides whether an on-image measurement is usable, and it has nothing to do with your camera.
Image magnification falls off with distance. If your camera is 1,000 mm from the reference plane, a feature that sits 100 mm closer to the camera images at 1000 ÷ 900 = 1.11× — 11% larger. Move the camera to 3,000 mm and the same 100 mm of depth costs 1000 ÷ 967 ≈ 3.4%. At 5,000 mm it is 2%.
On a 600 mm cabinet face with a 100 mm proud handle, that is the difference between a 66 mm error and a 12 mm error, from the same product, on the same day, with the same camera.
The rule: shoot from at least three times the longest dimension you intend to annotate, and use the longest focal length that still frames the product. Long lens plus long distance is how you approach an orthographic view, which is the only kind of photo you can safely measure.
Four Setups, Ranked by How Much They Lie
These are the four setups that actually get used in supplier photo rooms. The error magnitudes below are the typical published ranges for each optic class, not a claim about your specific device — measure yours once with a dot chart and you will know exactly.
| Setup | Typical geometric distortion | Perspective risk | Verdict for a spec image |
|---|---|---|---|
| Phone ultra-wide (0.5×), close in | Several percent barrel, strongest at the frame edge | Severe — the mode invites shooting close | Unusable for measurement. Fine for a mood shot |
| Phone main camera (1×) at ~1 m | Under a percent after in-phone correction | High at 1 m — 11% per 100 mm of depth | Only for objects with a flat, shallow face |
| Phone telephoto (2–3×) at 2.5–3 m | Low | Moderate | Workable if the camera is truly perpendicular |
| Camera at 85–100 mm equivalent, tripod, 3 m+ | Low, and a lens profile exists | Low | The one to standardise on |
Notice that the ranking is driven mostly by the last two columns. A modest camera at the right distance beats an expensive one held at arm's length, every time.
Getting the Camera Square
Perpendicularity is where most photo rooms quietly fail, because "it looks straight" is not a measurement.
- Put the camera axis normal to the face you are dimensioning — not to the floor, not to the product's centre of mass. If you are dimensioning the front, the lens points at the front, dead on.
- Set the lens height to the mid-height of that face. A camera set at table height photographing a 2,100 mm wardrobe is tilted by definition, and the top of the wardrobe will image narrower than the bottom.
- Level in two axes. Most phones have a built-in level; use it. A 2° tilt on a 600 mm face is roughly 20 mm of keystone across the frame.
- Correct the lens before you annotate, not after. Applying a lens profile after you have placed dimension callouts moves the pixels under the callouts.
Where the Scale Reference Goes
A ruler or a scale card in the photo only works if it sits in the same plane as the dimension it calibrates. A scale card leaning against the front of a sofa calibrates the front face of that sofa and nothing else. Put it on the seat, and every dimension you scale off the front face inherits the depth error.
Three practical placements:
- Flat face products (panels, doors, sheet goods): card taped to the face, in a corner, outside the annotated area.
- Deep products (sofas, machines, cabinets): one card per plane you intend to dimension, or accept that only the calibrated plane is measurable.
- Cylindrical parts (pipe, drums, bearings): card at the tangent plane touching the camera-facing surface, never against the back.
This is the part that makes the difference between an image that can carry numbers and one that just carries a mood, and it is why a spec sheet is built from a controlled shot rather than a catalogue hero image. Once the shot is right, the annotation should take its values from the measured product and lock them to the real edges, rather than from someone eyeballing pixel counts on a photo that was never geometrically honest in the first place. An AI-generated product image is the extreme version of the same failure: nothing in it was ever measured, so any dimension printed beside it is decoration.
For the downstream side of this — how the finished dimensioned image should be laid out and exported — dimension drawings for a product catalog covers the layout, and 300 dpi vs 72 dpi product images covers the resolution question that decides whether your callouts survive print.
Shot Checklist for a Measurable Photo
- Camera distance ≥ 3× the longest dimension being annotated
- Longest focal length that still frames the product (85–100 mm equivalent is the sweet spot)
- Ultra-wide / 0.5× mode switched off entirely
- Lens axis perpendicular to the face being dimensioned
- Lens height at the mid-height of that face
- Levelled in two axes, tripod or stable support
- Lens profile correction applied before any annotation is placed
- Scale reference in the same plane as the dimensions it calibrates
- One shot per plane if the product is deep
- Real measured dimensions recorded separately — the photo confirms them, it does not generate them
That last line is the one to argue about internally. A photograph is evidence for a dimension, never the source of it.
FAQ
How do I photograph a product for measurement?
Shoot from at least three times the longest dimension, with a long focal length (85–100 mm equivalent), camera axis perpendicular to the face being dimensioned and at that face's mid-height, with a scale reference in the same plane. Apply the lens correction profile before adding any callouts.
What focal length is best for product photography?
For measurable spec images, 85–100 mm equivalent on a tripod. Shorter focal lengths force you closer, which amplifies perspective error; much longer lengths need a studio depth most photo rooms do not have.
Why do dimensions measured off my photo not match the product?
Two likely causes. Either the camera was close enough that depth turned into scale error — at 1 m, 100 mm of depth is an 11% size difference — or the lens introduced barrel distortion that was never corrected. Both are geometry, not opinion.
Can I fix perspective distortion in software?
Partly. Keystone correction re-warps the image, which fixes converging verticals but stretches and softens pixels unevenly, and it cannot recover the depth-related scale error at all. Fixing it at the tripod is cheaper and more accurate than fixing it in software.
Does a phone camera work for spec photos?
The telephoto lens on a modern phone, on a tripod, at three metres, is usable. The main camera at arm's length is not, and the 0.5× ultra-wide should never be used for anything that will carry a dimension.