The KYC Photo Standard: The Specifications Automated Checks Actually Measure

Nearly every guide to photographing your ID tells you to make sure the image is clear. That is not a specification, it is a wish. Underneath the verification flow sit two published international standards with real numbers in them. This is what those numbers are, and how to hit them with a phone.

Last reviewed 30 July 2026 · by the KYC Rejected team

Why vague advice keeps failing you

When an automated verifier looks at your document it is not forming an impression. It is computing values and comparing them against thresholds. Sharpness is measured. Contrast is measured. The angle of the document relative to the sensor is estimated and corrected. The proportion of the frame covered by specular highlights is calculated. Each of these produces a number, and each number is either inside or outside a boundary.

That is why advice like take a clear photo in good light fails so many people. It describes an outcome without describing the mechanism, so when the photograph gets rejected you have no idea which of the dozen measured properties fell short. You retake it in a way that feels better to your eye, which is the wrong instrument, and you spend another attempt.

The useful move is to work backwards from the standards the industry actually builds against, because those tell you what is being measured.

The two standards underneath everything

ICAO Doc 9303 is the specification for machine readable travel documents, published by the International Civil Aviation Organization. It is the reason passports across the world share a common layout, and it defines the machine readable zone, the two lines of monospaced characters across the bottom of the photo page. It also specifies the optical character recognition characteristics those lines must have, including that security features printed over them must not interfere with reading the characters in the B900 range defined in ISO 1831.

ISO/IEC 19794-5 is the biometric standard for face image data. It sets out scene constraints such as lighting, pose and expression, photographic properties such as positioning and focus, and digital attributes such as resolution and image size. When a verification vendor tells you your selfie was unusable, this is broadly the vocabulary they are working in.

Neither standard was written for people photographing documents on a kitchen table with a phone. Both were written for controlled capture, at a passport office, with fixed lighting. That gap is precisely where most rejections live, and closing it is what the rest of this page is about.

Part one: the document image

Resolution, and what 200 dpi means on a phone

Dots per inch is a printing measurement, and it confuses people when applied to a digital capture, because a photograph does not have an inherent dpi until you decide how large the subject is. The practical translation is this: the reliability of machine readable zone recognition depends on how many pixels land across each printed character, and roughly 200 dpi of effective resolution across the document is the figure generally treated as the working minimum for readable optical character recognition with adequate contrast.

A passport photo page is about 125 millimetres wide, which is close to five inches. At 200 dpi that means you want on the order of a thousand pixels spanning the width of the document itself, not the width of the whole photograph. Any modern phone sensor produces vastly more than that, which is why the resolution of your camera is almost never the real problem.

The problem is how much of the frame the document occupies. If you stand back and capture your passport as a small rectangle in the middle of a large photograph of your table, the sensor may have twelve million pixels but only a small fraction of them are on the document. Fill the frame. The document should occupy most of the image, with roughly a centimetre of background visible on every side so the edge detector can find the outline.

One caution that surprises people: do not use digital zoom to fill the frame. Digital zoom crops and then interpolates, which invents pixels rather than capturing them. It produces an image that looks larger and contains no more real detail. Move the phone closer instead.

The machine readable zone is the highest value area in the frame

If you optimise one region of your document photograph, optimise those two lines of monospaced characters. They encode your name, document number, nationality, date of birth and expiry date, and they carry check digits, which means a reader that parses them successfully can validate its own reading. That gives the system a second, independent confirmation of your identity fields alongside the printed text.

The consequence is direct. A photograph with a small amount of glare across a decorative area may pass. The same amount of glare sitting across the machine readable zone will often fail, because it has destroyed the most reliable data on the page. When you check your shot before uploading, zoom into those two lines first and confirm every character is legible.

Glare is geometry, not brightness

The most common misunderstanding about glare is that it is caused by too much light. It is caused by the wrong angle. Modern identity documents are laminated, and laminate behaves like a mirror. Light striking it reflects away at the same angle it arrived. Glare appears in your photograph when a light source, the document surface and the camera lens happen to line up so the reflection travels straight into the sensor.

This explains why turning the flash on makes it dramatically worse. A flash sits within a couple of centimetres of the lens, so it is almost perfectly positioned to bounce straight back. It is close to the worst possible light source for photographing a laminated card, and it is also the one people reach for when a photograph looks too dark.

The fix has two parts. First, switch the flash off and move next to a window so the light arrives from the side rather than from behind the lens. Second, if a bright patch persists, break the alignment by tilting the document five to ten degrees. Slipping a coin under one edge is enough. Watch the highlight slide across the surface and stop when it is clear of all printed fields.

Holograms are the exception that confuses people. They are designed to catch light and will always show some interference. That is expected and is not automatically a failure, provided no printed field is obscured. Detailed walkthroughs live on our glare detected and hologram reflection pages.

Edges, perspective and why corners matter so much

Before a system reads anything it locates the document within the photograph, and it does that by finding four corners and the straight lines between them. Those corners let it estimate the plane of the document relative to the sensor and mathematically flatten the perspective, so that a page shot slightly from above becomes a clean rectangle for the reader.

Remove a corner and you remove the ability to do that correction. This is why cropping tightly to the edges of the document is counterproductive even though it feels tidy, and why a photograph taken with one corner hanging off the frame gets refused even when every word on the page is perfectly readable.

Contrast between the document and what it sits on matters for the same reason. A pale passport page on a white table gives the edge detector almost nothing to work with. A dark, plain, matte surface makes the boundary obvious. Avoid patterned surfaces, because a busy background produces false edges.

Hold the phone parallel to the document rather than leaning over it at an angle, and keep the whole page in frame with a margin. See document cropped for the failure mode in detail.

Colour, and why photocopies fail

Submit a colour photograph of the physical document. Black and white photocopies, scans converted to greyscale, and screenshots of a document displayed on another screen are refused routinely. Part of that is anti fraud logic, because colour and the behaviour of security features under light are among the signals used to judge whether the thing photographed was a real document. Part of it is simply detail, since converting to greyscale discards information the reader may have used. Photographing a screen adds its own problems, including moiré patterns from the pixel grid and the screen's own glare.

The compression trap

This one deserves its own section because it is invisible and it is common. If you photograph your document on your phone, send it to yourself through a messaging app, then download it on your laptop and upload that file, you have almost certainly ruined it. Chat platforms recompress images aggressively to save bandwidth. The version that arrives is smaller in file size and lower in real detail than the one you captured.

The damage lands exactly where it hurts most. Compression artefacts cluster around high contrast edges, and the machine readable zone is nothing but high contrast edges. To your eye the image still looks fine. To an optical character reader the characters have acquired a halo of noise, and the confidence score drops below threshold.

Upload the original file, straight from the camera roll on the device that took it. If you must move it between devices, use a method that preserves the file, such as a cloud drive or emailing it as an attachment rather than an inline image. Relevant background sits on our low resolution page.

Stop guessing what went wrong A human specialist corrects the document for you and sends it back ready to submit, usually in under 10 minutes. $1.99, money back if it is not approved. Get my fix

Part two: the face image

Distance between the eyes is the master measurement

Face recognition does not compare your selfie to your document portrait the way a person would. It builds a geometric description from landmarks, and the reference measurement underneath that description is the distance between your pupils. Everything else scales from it.

ISO/IEC 19794-5 puts the minimum inter eye distance at roughly 90 pixels, with about 120 pixels preferred. Below that the landmark positions become imprecise and the comparison loses reliability. The standard also frames it proportionally: the distance between the eyes should sit a little under a quarter of the image width.

In practice this means get closer, or let the camera get closer to you. Selfies taken at arm's length with your whole upper body in frame put far fewer pixels across your face than people assume. Fill a good portion of the frame with your head and shoulders.

Illumination has to be uniform, not merely sufficient

The requirement is not brightness. It is evenness. The standard asks for uniform illumination with no shadow falling across the face or across the background, no reflections, and no red eye. A single overhead bulb violates this immediately, because it carves a shadow under the brow and the nose, and those shadows sit precisely where the landmark detector is trying to work.

Front facing daylight is the easiest way to satisfy it. Stand facing a window, not with your back to it. Backlighting is a particularly reliable way to fail, because the camera exposes for the bright background and leaves your face underexposed, which then gets amplified with noise.

Remove anything that alters the geometry or occludes landmarks. Glasses cause reflections and can hide the eye region. Hats change the hairline. Masks and heavy scarves remove the jaw. Keep your whole head inside the frame, and make sure only one face is visible, because a second person in shot is a routine cause of refusal.

Pose, expression and focus

Bad illumination, head pose and loss of focus are the three factors most often responsible for disqualifying a face image. Face the camera squarely with a neutral expression. A tilted or rotated head changes the projected distances between landmarks, and while modern systems tolerate some rotation, you are spending tolerance you may need elsewhere. Hold still until the capture completes, since motion blur destroys landmark precision as effectively as being out of focus.

What a liveness check adds on top

A liveness check is a different problem from a face match. Its job is to establish that a physically present person is in front of the camera rather than a photograph, a screen, a printed mask or a replayed recording. That is why you are asked to turn your head, follow a dot, or move closer.

Most liveness failures come from moving too fast. The system needs to observe a continuous, natural transition, and a quick jerk of the head produces motion blur and skipped frames, which reads as suspicious rather than as compliant. Move slowly and smoothly, at roughly half the speed that feels natural.

If liveness keeps failing on a laptop, switch to your phone. Phone front cameras are usually far better than laptop webcams, and the difference in sensor quality alone resolves a large share of cases. Our step by step guide is at how to pass the liveness check, and the specific error is covered at liveness check failed.

A capture procedure that satisfies both standards

Run this in order. It takes about five minutes and removes most of the measurable reasons for refusal.

1. Choose daylight. Position yourself beside a window, not in front of it and not under a single ceiling light.
2. Switch the flash off, and confirm it is off rather than on automatic.
3. Clean the lens. A fingerprint smear reduces contrast across the whole frame and is the cheapest fix available.
4. Lay the document on a dark, plain, matte surface.
5. Brace the phone. Rest your elbows on the table or prop the phone against a stack of books. Handheld capture at close range is the main source of subtle blur.
6. Hold the phone parallel to the document, not angled over it.
7. Fill the frame, leaving about a centimetre of background on all four sides. Do not use digital zoom.
8. Tap the screen directly on the document to lock focus and exposure there.
9. Take three or four shots, shifting the tilt slightly between them so at least one has no highlight over any printed field.
10. Zoom in and inspect. Check the machine readable zone character by character, then the date of birth, then the document number.
11. Repeat for the reverse side if the document is a card.
12. For the selfie, face a window, remove glasses and hats, fill the frame with your head and shoulders, hold still, and move slowly through any prompted motion.
13. Upload the original files from the device that captured them. Do not route them through a chat app.

Our guide to taking perfect ID photos goes deeper on the photography itself.

What the standards do not cover

It is worth being clear about the limits. Hitting every number above guarantees only that your image is technically adequate. It does not guarantee approval, because a significant share of rejections are not image problems at all. A name that does not match your profile, an expired document, a jurisdiction the platform does not serve, or a fraud model that has linked your device to an existing account will all refuse a technically flawless photograph. Those categories are covered on our error code index. Photography solves the photography problems, and nothing else.

Frequently asked questions