DocSupra › Fix a blurry document photo · Español
How to fix a blurry photo of a document
DocSupraCompliance Studio
Fix this file now
Drop the document here: it opens in the Compliance Studio, which asks where it is going. Nothing is uploaded.
Your browser reads the file on this device: DocSupra never uploads it. You see the finished file before you pay; paying is only for the download. Terms · Privacy
Most rejected document uploads are not blurry in the way people think. They are shadowed, angled or glare-damaged — and those are all recoverable. Genuine motion blur is not. Here is how to tell which one you have.
Four problems, four different fixes
Shadow across the page
Usually your own head or the phone blocking an overhead light. The page ends up bright at one edge and grey at the other. This is the easiest to fix: estimate the lighting across the sheet and divide it out, and the page comes back to even white with the text intact. Nothing is lost, because the text was never gone — it was just sitting on a gradient.
Flash glare
A bright blown-out blob, usually on a laminated card or glossy paper. Partly recoverable. The surrounding area can be normalised so the glare does not turn into a black blob when the image is converted to black and white. But where the glare completely blew out the surface, the text underneath is gone — no software recovers it. If the glare sits on the data you need, retake the photo without flash.
Taken at an angle
Fully recoverable. Detect the four corners of the document, compute the transform that maps them back to a rectangle, and resample. The result looks like a flatbed scan. This also fixes the trapezoid distortion that makes the far edge of a document look smaller than the near edge.
Genuine soft focus
Only partly recoverable. Sharpening restores the appearance of edge definition, but detail that the sensor never captured cannot be invented. If the small print was never resolved, it stays unresolved. Anything claiming to fully restore focus is guessing at the missing detail.
Be sceptical of "AI enhancement" for documents. A model that invents plausible detail is fine for a holiday photo and dangerous on a tax form, where an invented digit is worse than a blurry one. Corrections to lighting, angle and contrast are measurable and reversible. Generated detail is neither.
How to photograph a document so you only do it once
- Indirect daylight, no flash. Near a window, not in direct sun. Flash is the single biggest cause of unusable document photos.
- Fill the frame. Get close enough that the document occupies most of the shot. This is what determines whether small print survives.
- Shoot straight down. Phone parallel to the page, not tilted.
- Watch your own shadow. Stand to the side of the light, not between it and the page.
- Flatten folds first. A crease throws a shadow line that reads as a mark on the document.
- Plain contrasting background. A dark desk under white paper helps edge detection find the document.
Fix shadow, glare and angle in about a second
Runs in your browser. Your document is never uploaded. See the finished result before you pay.
Open DocSupraYour phone exposed for the room. The document came out grey.
The short answer. A camera tries to make the average of what it sees middle grey. Point it at a white page filling the frame and it darkens everything to compensate — your white paper becomes grey and the black ink becomes dark grey. The information is still there; the contrast is not. Recovering it is a tone problem, and tone is recoverable in a way that lost pixels are not.
It looked dim on the screen and you took it anyway because the light was not going to get better. Then the portal said it was not legible.
The 5 reasons, in plain words
- The camera metered the wrong thing. Filling the frame with white paper is exactly the case where automatic exposure gets it wrong, and it gets it wrong in the same direction every time: too dark.
- Your own shadow. Leaning over a page to photograph it puts your head between the light and the paper. The darkest part of the image is usually the part you were reading.
- Grey paper fails legibility checks. A reviewer or an automated check compares ink against paper. When the paper is grey, that difference collapses and the page reads as unclear even though it is not.
- Brightening the whole image does not work. Raising everything equally lifts the ink along with the paper, and the difference between them stays exactly as small as it was. It is contrast that is missing.
- Dark and blurry are different failures. A dark photo holds its detail and can be recovered. A blurred one does not, and no amount of tone work will put back an edge the lens never resolved.
Which of them can software fix?
The tone, which is the part that is genuinely recoverable. The paper is measured block by block and lifted to a clean white while the ink is held down, so the contrast comes back without the text thickening or breaking up. If the capture was also out of focus, that is said plainly — it is the one thing that cannot be repaired after the fact.
The flash left a white hole in the middle of your document.
The short answer. A flash fired straight at a flat sheet bounces straight back. Where it lands, the sensor records the maximum value it can and stops — there is no darker detail stored underneath, because the light overwhelmed it. Glare repair recovers the areas that are merely bright. Where the sensor genuinely saturated, nothing can bring back what was never recorded, and you will be told so instead of being sold a rescue.
It is the most frustrating kind of bad photo, because the document was perfectly fine and the phone was trying to help.
The 5 reasons, in plain words
- Paper is more reflective than it looks. A sheet of office paper bounces a large share of the light that hits it. Pointed straight at a flash, it behaves closer to a mirror than to a wall.
- The angle decides everything. Glare appears where the angle from the flash to the paper equals the angle from the paper to the lens. Moving twenty centimetres to one side usually removes it.
- Bright and saturated are different. A bright area still holds faint variations that can be pulled back. A saturated one is a flat maximum — every pixel identical, nothing underneath to recover.
- Laminated documents are the worst case. A licence or an ID card in plastic reflects far more sharply than paper, which is why the glare lands as a hard-edged patch instead of a soft one.
- Turning the flash off is usually enough. Indirect daylight near a window, no flash, is almost always better than a flash at any angle — and it costs nothing to try before you fix anything.
Which of them can software fix?
The recoverable part, and only that. The bright zone is measured against the paper around it and pulled back to match, so the page reads evenly. Where the sensor saturated, the result says so rather than inventing detail — a document with invented content is worse than a document with an honest hole in it.
Removing the shadow from a photo of a document
The short answer. Move the page so the light comes from in front of you rather than behind, or use a window instead of a ceiling light. If the photo is already taken, the shadow can be removed by rebuilding the tone across the page — but only if the shadowed area still holds detail.
You lean over the page to line up the shot, and the light you are working under is behind you. The shadow is not bad luck — it is geometry, and it happens to everybody.
Why the shadow lands exactly on the page
Indoor light almost always comes from above and slightly behind where you stand. To photograph a page on a table you lean over it, putting your head and shoulders between the lamp and the paper. The darkest part of the shadow falls where you are looking — which is the middle of the document.
Preventing it, which is far easier than fixing it
- Use a window, not the ceiling light. Put the page on a table beside a window with the light arriving side-on.
- Stand so the light is in front of you. If you can see your own shadow on the page, turn around.
- Do not use flash. Flash replaces a soft shadow with a hard white blowout over whatever it hits, and a blown-out area has no detail left to recover.
- Shoot from directly above. A tilted camera puts one edge of the page further from the light than the other, which adds a gradient on top of the shadow.
Removing one that is already there
A shadow is a smooth change in brightness across the page. The text underneath it is still there — just darker. Correcting it means measuring the paper's brightness in each region and lifting each region to the same white, so the paper reads uniformly white and the ink reads uniformly black.
What cannot be recovered is a region that clipped. If part of the page went fully black, or the flash blew part of it to pure white, there is nothing left in those pixels to lift. That part has to be rephotographed.
Beware software that "fixes" this by thresholding. Forcing every pixel to either black or white makes a shadow vanish instantly — and takes the periods, commas and thin accents with it, because those are grey at the edges. The page looks clean and a date reads wrong.
The document is straight. The photo is not.
The short answer. Photographed from one side, a rectangle becomes a trapezoid: the near edge longer than the far edge, lines of text that converge, letters that shrink across the page. That is perspective, not rotation, and turning the image by a few degrees does not undo it. The page has to be found by its four corners and pulled back into a rectangle.
Nobody holds a phone exactly above a page. The light is in the way, or the shadow of the phone is, so the photo is taken from a little to the side. The eye forgives it. A checker does not.
The 5 reasons, in plain words
- Perspective is not rotation. A tilted photo is fixed by turning it. A photo taken from the side is a different shape, and no rotation makes a trapezoid a rectangle.
- Text that converges fails the legibility check. Lines that run together towards the far edge read as smaller type. The far part of the page can fall below the size a reviewer accepts.
- The edge is the clue. A card or a sheet of paper has four straight edges. When they can be found, the correction is exact: the corners say where the rectangle was.
- What the camera did not see is not there. A page photographed from a steep angle has fewer pixels at the far end. Straightening spreads them out; it cannot add detail that was never captured.
- Shadows ride along. Side light means a shadow across the page. It is a separate problem, fixed in the same pass, but it is not perspective.
Which of them can software fix?
The straightening, when the edge can be found. The engine looks for the four edges of the card or the page; with them found, it pulls the document back into a rectangle, trims the table or the desk around it, and for a page cleans the paper to an even white with the ink untouched. It tells you when it could not find the edge, and then it changes nothing rather than guess. Blur and a corner outside the frame are not fixable, and it says so. In this browser, with nothing uploaded.
The scanner pulled the sheet in sideways. The page came out tilted.
The short answer. A flatbed or a document feeder almost never receives the sheet perfectly aligned: two or three degrees are enough for the lines of a form to run diagonally across the page and for the portal's automatic reader to start failing. Straightening is not rescanning: the angle of the sheet is measured, the image is rotated by exactly that angle, trimmed to the paper and exported again at the format, resolution and size the destination asks for. The content does not change; the geometry does.
It shows most in what comes out of the shared scanner at the office, the pharmacy or the post office: the sheet goes in turned, the lid leaves a grey wedge along the edge, and the PDF you are emailed is already leaning.
Why a scan comes out crooked, and why it matters
- The scanner guide does not hold the sheet. A document feeder accepts a sheet even when it enters two degrees turned, and on a flatbed a sheet placed by eye against the corner is rarely square.
- A crooked scan is a rotation, not a perspective. Unlike a phone photo, the four corners keep their right angles: the whole sheet is turned on the glass. That is why the fix is a measured rotation rather than a stretch, and why nothing is lost or deformed.
- The automatic reader fails before the eye does. A portal that extracts data from the document follows the text lines; with the page tilted it mixes lines and columns and flags the file as "not legible" even though a person reads it without effort.
- The wedge at the edge adds weight and draws the eye. When the sheet goes in turned, the scanner captures a triangle of the lid background, grey or black. That wedge adds kilobytes to the file and is the first thing a reviewer sees.
- A multi-page scanned PDF is not straightened in one go. Each page may have entered at a different angle. DocSupra prepares and exports page 1 of a PDF: if the scan has several pages, split the PDF first or scan each sheet as an image and correct them one by one.
What DocSupra does with a crooked scan
It finds the edge of the paper, measures the angle, rotates the page until it is square and trims the background wedge. Then it exports it the way the destination asks: JPEG, PNG, PDF or PDF/A, at 300 DPI when the destination requires it, in black and white if it is a text-only page, and under the size cap. All of it happens in your browser: the scan is not uploaded to any server, you see the finished result before paying, and paying is what downloads it.
What it does not fix. A page curved by the spine of a book, text that fell outside the glass, or a strip of the lid that covered part of the content. That is not in the file, and it has to be rescanned.
Questions
What is the difference between a crooked scan and a photo taken at an angle?
A crooked scan is rotated: its corners are still square and the whole sheet turned a few degrees on the glass or in the feeder. A photo at an angle is in perspective: one edge comes out shorter than the other. The same corner correction handles both; on a scan it reduces to a measured rotation and the page is not stretched.
Can I straighten a multi-page scanned PDF?
DocSupra prepares and exports page 1 of a PDF. If the scan has several pages, split it first or scan each sheet as an image: each page may have entered at a different angle, and they are best corrected one by one.
Does straightening lower the quality of the text?
A rotation of a few degrees is a single resample of the image, and the text stays as sharp as the scanner captured it. DocSupra also measures the sharpness of the final file: a blurry result is not delivered.
What resolution should I scan at so the file is not sent back?
300 DPI is the reference most portals ask for on text documents; below 200 DPI small print starts to break up. Scan at 300, straighten, and bring the size down afterwards with the format and the compression, never with the resolution.
Related
Doing this for clients, not for yourself? DocSupra for teams adds the batch queue, one combined PDF per matter and a dated compliance report — still with nothing uploaded.
What it costs
See the finished file first — sharp on screen, 100% with the lens, no card and no account.
Prices are in USD and exclude any tax that applies in your country; Stripe calculates it and shows it before you pay.
See the full comparison →