مسح ضوئي إلى PDF

ماذا تفعل هذه الأداة

Most people do not own a scanner and everybody owns a camera, but a photograph of a page is not a scan. This closes that gap in four steps: the page is found in the frame, the perspective is flattened, the text is straightened, and the image is thresholded so it reads as a document rather than as a picture of one. On a phone the camera panel captures directly; on a desktop it falls back to the file picker, because most people there are dropping photos they already took.

Use it when something has to be submitted as a scan and there is no scanner in the building: an identity document or a certificate for an application, a signed page going back to an office, a receipt for an expenses claim, or a handwritten note that has to be filed with everything else.

كيف تعمل

1. Take the photo here, or drop photos you have already taken. Up to 100 at a time, each up to 100 MB, and every photo becomes one page in the order they are listed.
2. Leave find the page and straighten on. A detection the tool is not confident about is discarded rather than guessed at, and the whole frame is used instead.
3. Choose the mode. Black and white is the default and is what makes a photograph look like a scan; grayscale suits a page with photographs on it; colour keeps a stamp or a signature in its own ink.
4. Pick the page size. A4 or Letter puts every photo on a standard sheet; Match makes each page the shape of its own photo.
5. Turn on the searchable-text option if you will need to find or copy words later, and pick the language.
6. Press Scan. The pages are assembled into a single PDF and the download starts.

The four steps run in that order because each depends on the one before. The document is a bright quadrilateral somewhere in a frame that also contains a desk, so its outline is detected first. A homography then maps that trapezoid back to a rectangle, which is what removes the leaning look of a photo taken at an angle. Only then is the residual rotation measured, and it is measured from the text lines rather than the page edges, because lines of text give away the true angle more reliably than a torn or shadowed edge does.

Thresholding is the step that decides whether the result is usable. A photograph has uneven lighting, so a single global threshold turns one half of the page black and blows the other half out. An adaptive threshold compares each pixel to its own neighbourhood instead, which is why a phone photo comes out looking like a scan rather than like a photocopy of a shadow. It is also why black and white is the default rather than an option for the brave.

The mode chooses the encoding as well as the look, and that is where the file size goes. A thresholded page is two colours and compresses to almost nothing as PNG, while JPEG would put ringing artefacts around every letter. A grayscale or colour page is photographic and is the other way round, so those are written as JPEG at quality 0.85. A black-and-white scan is routinely several times smaller than the same page in colour.

Resolution is set in dpi and defaults to 200, which is enough for ordinary printed text. Raise it to 300 for small print, faint carbon copies or anything the recognition is struggling with. 400 stores four times the pixels of 200 and rarely reveals something the threshold at 200 had not already found, so it is worth checking one page before committing a hundred. EXIF orientation is read and applied before any of the geometry, so a photo your phone tagged sideways is not straightened into the wrong rectangle.

Recognition, when you switch it on, runs on the processed pages rather than the originals - that is the whole point of doing the image work first, because a straightened, thresholded page recognises far better than a photograph of one. The words are placed as an invisible text layer over the image, so the page looks exactly as it did and the text can be selected, searched and copied. Words the recogniser scored below 40 per cent are left out rather than added wrong, since a wrong word in a searchable document is worse than a missing one.