Phone-Scan Deskew to PDF

Why a phone photo of a page looks crooked

Perspective distortion, not blur, is what makes a phone photo of a document unreadable as a scan. Point a camera at a flat page from any angle other than dead-on, and the edges nearest the lens project larger than the edges farther away, turning a rectangle into a trapezoid. Photographers call this keystoning, and it happens even when the photo itself is perfectly sharp and well lit.¹

A flatbed scanner avoids the problem entirely by holding the page and the sensor at a fixed, known distance across the whole surface. Your phone has no such fixture, so the page's four corners land at four different distances from the lens every time you lift the camera by hand. Straightening the result afterward means finding those four corners and mapping them back onto a true rectangle, which is exactly the geometry problem this tool solves.

Every phone camera saves its photos in whatever format is set as default, and this tool accepts JPEG, PNG, WebP, HEIC, or HEIF without you needing to convert anything first. HEIC is what an iPhone shoots by default.² Inside the tool, decoding a HEIC file works the same way as every other format: through a small WebAssembly decoder running in your browser tab, so an iPhone photo never needs a server round trip just to open it.

How four corners become a straight rectangle

Once you mark where the page's four corners actually sit in the photo, the tool solves a projective transform, a well established piece of computer-vision math that answers one question: what mapping turns this exact quadrilateral into a rectangle, and what does every pixel inside it look like once that mapping is applied. The four points you drag are enough information to solve it completely, because a projective transform in two dimensions has exactly eight degrees of freedom and four point correspondences supply exactly eight constraints.³

From that solved mapping, the tool works backward from every pixel in the finished rectangle to the exact fractional position it came from in your original photo, then blends the four nearest source pixels together for a smooth result instead of a blocky one.⁴ The output rectangle's width and height come directly from the average length of the top and bottom edges and the average length of the left and right edges you drew, so a portrait letter page and a landscape receipt both end up close to their real proportions without you choosing a paper size first.

Why manual corners beat a guess

Automatic page-edge detection works by looking for contrast between the document and whatever is behind it, and it fails in exactly the situations phone scanning most often happens in: a wood table with a similar tone to the paper, a cluttered desk, or window glare across one edge. Dragging the corners yourself removes that guesswork, at the cost of a few seconds of setup per page.

Precision matters more than speed at this step, so the four handles support more than a raw drag. Each one also answers arrow keys, nudging a few image pixels per press and a larger step when you hold Shift, which helps when a corner needs a correction finer than a fingertip on a small screen can reliably place. Zooming in before that final adjustment sharpens the same nudge further, since the handle's step size stays fixed in image pixels while the photo underneath grows larger on screen.

Original, Grayscale, and B&W

Once a page is straight, the three enhance presets decide how its colors are rendered before it joins your document. Original keeps the straightened photo's colors exactly as captured, which matters when a document carries a colored stamp, a highlighted line, or a photo. Grayscale drops color entirely but keeps smooth shading between tones, closer to how a black-and-white photocopier renders a page.

B&W goes further, converting every pixel to pure black or pure white using a threshold that adapts to the local brightness around it rather than one fixed cutoff for the whole page.⁵ Because the threshold is local, B&W stays readable across a page with uneven lighting, such as a slight shadow from your hand near one edge, in a way that a single global cutoff would not. The result is also the smallest file of the three, since large flat regions of pure black or pure white compress far better than the subtle gradients a photo or a grayscale scan retains.⁶

Skipping the PDF for a single image

You don't have to build a PDF to get value from this stage. Download PNG on the enhance panel saves just that one corrected image in whichever preset you have selected, which suits a single receipt or ID page destined for an email attachment or an upload field that wants a picture rather than a document.

That download stays lossless, so the pixels reaching your disk match what the enhance step produced with nothing softened or recompressed around sharp edges.⁷ For the B&W preset especially, every edge is already pure black or pure white, and a lossy format would risk blurring that crisp cutoff back toward gray. The file arrives named with today's date under the tool's own prefix, so a handful of single-page downloads stay easy to tell apart later.

Building a multi-page document from separate photos

A lease, a set of homework sheets, or a multi-page utility bill rarely gets photographed in one shot, so the tool treats every photo as its own page with its own corners and its own enhance mode. Add one, straighten it, choose how it should look, and it joins a page stack shown as a row of thumbnails; add the next page the same way, and repeat until every page of the source document is represented.

Fixing a page that came out wrong

Because paper shuffles and phones get held at different angles between shots, pages rarely arrive in the right order or need identical treatment. The stack's reorder controls fix sequencing after the fact, and re-opening any page's pencil icon returns you to that page's saved corners and mode so you can adjust just the one page that came out wrong, rather than starting the whole document over.

Beyond the up and down buttons, every thumbnail in the stack is itself draggable, so picking one up and dropping it between two others reorders the whole stack in a single motion once you have more than a page or two to sort. An icon on the same thumbnail removes a page outright, which matters when a blurry duplicate or a page you photographed twice by accident ends up sitting in the stack.

Searchable text without leaving your device

A photographed page is just pixels to a PDF reader until something adds a text layer, which is why archived scans are often unsearchable even when every word on them is perfectly legible to a person. Checking Make this PDF searchable runs a local text-recognition engine, the same one that powers this site's OCR Redactor tool, against each finished page, right inside your browser tab.

The recognized words are drawn back onto the page at the exact position and size they occupy in the image, but invisibly, so what you see is still the scanned photo while what a PDF viewer can select and search is the recognized text sitting exactly on top of it. That invisible-layer approach is the same technique long used by dedicated OCR-to-PDF software, and it means a page recognized this way stays visually identical to the same page without OCR while gaining full-text search in whatever tool you archive it with.⁸