Scan to PDF Turn a camera into a document scanner, without installing anything

How to Scan a Document to PDF with Your Phone or Webcam

Three minutes for a short document, nothing installed, and no account for your first run. This document scanner runs inside the page you are reading, using your device camera and your own processor.

1. 1

   Start the camera, or choose photos you already took

   Press Start camera and allow access when the browser asks. The rear camera is picked automatically on a phone, because it is the better of the two. Some machines have none. Then Choose photos accepts pictures you have already taken, and the code under Use my phone instead opens this same scanner on a phone.

2. 2

   Pick the look before you shoot

   Enhanced stretches the brightness so paper reads as white, and it is right for most documents. Black and white applies an adaptive threshold and gives the smallest, sharpest, most scan-like result for plain printed text. Grayscale keeps the tones for photographs and diagrams. Untouched leaves the capture alone. The setting applies to the next capture, never to pages you already took.

3. 3

   Fill the frame and press the shutter

   Hold the camera square over the page rather than at an angle, get the paper as close to filling the frame as you can, and keep your own shadow off it. Every capture becomes one page, and the board fills up underneath as you go.

4. 4

   Fix the pages, then build the PDF

   Drag pages into order, turn any that came out sideways, delete the bad shots. Pick a paper size. Press Build PDF, and the document is assembled on your own processor and downloads immediately, with no upload step anywhere in between.

The dashed rectangle in the viewfinder is a framing guide rather than a crop. The capture is the whole frame, exactly as you see it. Line the paper up inside the guide and you get a page with a sensible margin and nothing important sliced off at the edge. Nothing is thrown away.

Using Your Phone as the Scanner for a Desktop Computer

The desktop webcam pointing at your face is a bad document scanner. It is 720p, it is fixed above the screen, and the only way to photograph a sheet of paper with it involves holding the page up and reading it backwards. The phone in your pocket has a far better camera and it can be pointed downwards. So the question everybody arrives with is: how do I get the phone involved?

There are exactly two honest answers, and the panel above offers both.

Route one: your phone as a camera on this machine

Modern operating systems already present a phone as an ordinary video device. Continuity Camera on macOS shows a nearby iPhone in every camera list, including the one this browser puts in front of you, and Phone Link does the equivalent on Windows 11 with an Android handset. Any USB or wifi webcam app does the same on any platform.

Once the operating system sees the phone as a camera, this tool sees it too. Press Start camera, then pick the phone from the device list. You capture with the phone, you watch the viewfinder and the page board on the big screen, and the PDF is built on the computer. This is the best of both.

Route two: your phone as the whole tool

Open the handoff card and point your phone camera at the code. It opens this same scanner on the phone, where the rear camera is almost certainly the best one you own. You photograph the document there, and the finished PDF is saved on the phone, ready to email, upload or drop into cloud storage. No second device is involved.

Why the photos cannot be sent to your desktop

This is where every other tool gets vague, so here it is plainly: a web page on a phone cannot hand a file to a web page on a desktop without a server in between. The two devices have no direct channel. Every product that offers this feature is relaying your document photographs through infrastructure it operates, or brokering a peer connection through a signaling server that still sees who is talking to whom.

That is a legitimate engineering decision and it is incompatible with the promise printed on every page of this suite. So we do not offer it, and we do not pretend the QR code does something it does not. The code opens the tool on your phone. What happens next happens on your phone.

And if you have a real scanner

A flatbed or a document feeder produces image files in a folder, and browsers have no access to scanner hardware at all - there is no web standard for it and there is unlikely to be one. Scan to a folder as JPG or PNG, then press Choose photos and drop the whole folder in. The pages are processed exactly as camera captures are, and the ordering board works the same way.

The Engineering Behind Turning a Photograph into a Scan

A photograph of a page and a scan of a page are not the same object, and the difference is not resolution. A flatbed scanner lights the paper evenly from a fixed distance. A camera photographs a page lit by a window on one side and a lamp on the other, with the device casting a shadow across a corner. The raw capture is a gray-beige gradient with a dark patch in it, and it reads as a photograph of a document rather than as a document.

Three corrections close most of that gap, and each one is a control above.

Enhanced: a percentile brightness stretch

White paper in a phone photograph is rarely above 70 percent brightness, and the ink is rarely below 25 percent. The whole image occupies the middle of the range and looks washed out.

Enhanced builds a brightness histogram of the capture, finds the value below which the darkest half a percent of pixels fall and the value above which the brightest half a percent fall, and stretches that span across the full range. Percentiles rather than the true minimum and maximum, and that detail is what makes it work: one dust speck at pure black and one specular reflection off a staple would pin the range to 0 and 255 and the correction would do nothing at all. Ignoring half a percent at each end ignores both without touching any real ink.

A mild gamma lift is applied on top, which brightens the paper more than it brightens the text. The result is a color photograph with white paper, which is what you want when the document has a colored logo, a highlighter mark or a signature in blue ink worth keeping.

Black and white: an adaptive threshold, not a contrast slider

This is the setting that produces a real scan, and it is worth understanding why it beats the obvious alternative.

The obvious approach is a single cutoff: every pixel darker than some value becomes black, everything else becomes white. This fails on every real document photograph, because one side of the page is always brighter than the other. Any cutoff dark enough to keep the text in the shadowed corner turns the sunlit half into a solid black slab, and any cutoff bright enough to keep the sunlit half white erases the text in the shadow.

The fix is to compare each pixel against the average brightness of its own neighborhood instead of against a constant. A pixel that is darker than the paper immediately around it is ink, wherever on the page it sits and whatever the lighting is doing there. This is Bradley and Roth's adaptive threshold, and it is essentially what a flatbed scanner's text mode does in firmware.

Computing a local average directly would mean, for every pixel, summing every pixel in a window around it - on a 3500 pixel capture with a 400 pixel window that is a number nobody wants to write down. Instead a summed area table is built in one pass, after which the total of any rectangle is four array lookups regardless of its size. The whole operation is two passes over the image and finishes before you have let go of the shutter.

Bilevel output is written as PNG rather than JPEG, deliberately. JPEG is at its worst on pure black text against pure white paper: every letter edge picks up a halo of ringing, and the file is usually larger than the PNG, because deflate compresses long runs of identical pixels extremely well. A black and white page is frequently a tenth the size of the same page in color.

Resolution: the long edge cap

A 4K frame spread across the long edge of an A4 sheet works out at about 330 dpi, and a 48 megapixel phone still pushes it past 600. Both sit above the 300 dpi that document archives standardize on, and above what the eye resolves on paper at reading distance. The extra pixels buy you file size. Nothing else.

So each capture is resampled to the long edge that matches the resolution you chose, measured against A4:

How to Photograph a Page So It Comes Out Like a Scan

The processing above fixes lighting. It cannot fix framing, focus or a page photographed at an angle, and thirty seconds of technique beats any amount of correction afterwards.

1. Get square over the page. Hold the camera parallel to the paper, directly above the middle of it. A page shot from an angle comes out keystoned, wider at one end than the other, and nothing in this tool corrects perspective. This is the single biggest factor.
2. Fill the frame. Every pixel spent on the table around the page is a pixel not spent on the text. Move closer until the paper nearly touches the framing guide.
3. Keep your shadow off the paper. Shoot with the light coming from behind the camera or from the side, never with your own body between the window and the page.
4. Use a flat, plain surface. A wooden table is fine. A patterned tablecloth is not, and a lap is worst of all because the page curves.
5. Flatten a folded document first. A crease throws a shadow line that the threshold reads as ink, and a fold running through a paragraph is the one thing the black and white mode handles badly.
6. Let the camera focus. Tap the page on a phone screen and wait for the focus to settle. A slightly out of focus capture at 300 dpi is worse than a sharp one at 150.
7. Avoid the flash on glossy paper. It produces a white hotspot that no correction recovers. Move to a window instead. The torch button, where the browser supports it, is for genuinely dark rooms and matte paper.

Your Camera Feed and Your Documents Stay on This Device

This tool asks for permission to use your camera, which is the most invasive permission a web page can request. It is worth being precise about what happens with it.

The video stream is attached to a viewfinder element on this page and nothing else. When you press the shutter, the current frame is drawn onto a canvas in your browser's memory, processed there, and added to the board. No frame is transmitted, no still is uploaded, and the assembled PDF is built on your own processor. There is no server-side step anywhere. That is why the whole thing works with the network disconnected.

Two smaller things the tool does because they are the right defaults rather than because anybody asked:

• The camera stops when you switch tabs. A recording light that stays on while you are reading email is both a battery drain and, quite reasonably, the kind of thing that makes people distrust a web page.
• Photograph metadata does not reach the PDF. A capture is drawn from pixels and encoded fresh, and a photo you choose from your device is decoded and re-encoded the same way. Any GPS coordinates, camera serial number or timestamp your phone embedded are simply not carried into the document.

The honest trade-off, stated as it is on every tool here: because the work happens on your device, a twenty page batch at full camera resolution is limited by your phone rather than by our hardware. A server-side scanner would be more consistent on an old handset. It would also have your passport.

What This Scanner Will Not Do

The features a commercial scanner app has that this does not, stated plainly rather than discovered halfway through a twenty page document.

• It will not detect the edges of the page. The capture is the whole frame. Fill it properly and you will not miss the automatic crop.
• It will not correct perspective. A page photographed at an angle stays keystoned. Getting square over the paper takes two seconds and produces a better result than any correction, because a de-skewed capture is interpolated pixels rather than real ones.
• It will not straighten a slightly rotated page. Quarter turns only, on the board. Automatic deskew needs edge detection, which needs the edges, which is the first limitation again.
• It will not make the text searchable. The output is a picture of a page, so Ctrl+F finds nothing in it. Run OCR PDF on the result to weld a searchable text layer onto it.
• It will not scan from a flatbed or a feeder. No browser has access to scanner hardware, on any platform. Scan to a folder and use Choose photos.
• It will not send captures between two devices. That requires a server, explained in full above.
• It will not read camera RAW. RAW is sensor data rather than an image, and no browser decodes it. JPG, PNG, WebP, GIF, BMP and, on the platforms that support them, HEIC and TIFF are all accepted.

The batch ceiling is twenty pages. That is a memory decision rather than a policy one: a processed capture at 300 dpi is several megabytes, and twenty of them held in a phone tab alongside the PDF being assembled is already ambitious. For a longer document, build it in two or three parts and combine them with Merge PDF.

Common Scanning Problems, and the Fix

Frequently Asked Questions

The rest of the PDF suite

All of these run the same way: your file is read by JavaScript on your own machine, the work happens on your processor, and the result is handed back to you as a download that never existed anywhere else.