This document scanner turns phone photos of receipts, invoices, forms and book pages into a clean, flat PDF. Drop one photo or a whole folder and it finds each page's edges automatically, either with the instant built-in detector or with OpenCV.js, which you can download once for tougher photos, then straightens the perspective so a page shot at an angle comes out as a proper rectangle. If it picks the wrong edge, drag the four corners onto the page yourself and watch the result update beside it. Choose enhanced color, grayscale or a shadow-free black and white that makes text crisp, rotate any page, reorder them, and export everything as one multi-page PDF on A4 or Letter, or as a ZIP of images for your other PDF tools. It is free, works entirely in your browser, and your documents are never uploaded, which matters for bank statements, IDs and contracts. For best results shoot the whole page on a darker surface without glare.
| Size | Millimeters | Inches | PDF points |
|---|---|---|---|
| A4 | 210 × 297 | 8.27 × 11.69 | 595 × 842 |
| US Letter | 216 × 279 | 8.5 × 11 | 612 × 792 |
| Same as photo | — | — | Photo size at 150 dpi |
Photograph each page, drop the photos here, check the detected edges, then click Create PDF. The pages are flattened, cleaned up and combined in order.
Drag the four corner handles onto the corners of the page. The preview beside it updates as you move them.
Black & white for text pages, as it removes shadows and paper tint; Enhanced color for receipts with stamps, photos or colored forms.
No. Everything, including edge detection and PDF creation, runs in your browser.
The built-in engine is instant and works well for a page on a darker surface. OpenCV.js is a one-time download of about 13 MB that finds page outlines from edges, so it copes better with light desks and busy backgrounds. You can remove it any time with Clear downloaded models.
Only in Safari, which can decode HEIC. In other browsers, export the photos as JPG first.
Each photo is decoded with its camera orientation applied. A 400-pixel grayscale copy is thresholded with Otsu's method, the largest bright connected region is taken as the page, and its four corners are the points that maximise x + y, x − y and their opposites; if no page-like region covers at least 15% of the frame, the whole photo is used and the corners can be dragged by hand. The page is flattened with a perspective transform: a homography is solved from the four corners to a rectangle sized from the page's longer opposite edges, and every output pixel is sampled from the photo with bilinear interpolation. Enhanced color stretches each channel's 1st–99th percentiles to full range; black and white uses an adaptive local-mean threshold from an integral image, which removes shadows. Visitors can instead choose the OpenCV.js engine: the official OpenCV 5.0 build is downloaded once on request from jsDelivr, verified against a pinned SHA-384 fingerprint before it runs, cached in the browser and removable from the Downloaded models panel. With it, pages are found by Gaussian blur, Canny edges, dilation and contour search for the largest convex four-sided outline (falling back to the minimum-area rectangle), and flattened with cv.warpPerspective using bicubic interpolation. Pages are JPEG-encoded and placed on A4, US Letter or same-size PDF pages with pdf-lib.