Scan to PDF

Why a photograph of a page looks wrong

A scanner lights the paper evenly and sees the ink against it. A camera sees whatever light is in the room, so a white page arrives as a grey gradient with the darkest corner at the edge of the frame. There is no way to relight it, but there is a way to stretch the tones: the tool measures how dark and how bright the pixels really are, ignores the worst half a percent at each end so one dark corner cannot spoil the measurement, and rescales everything between so paper becomes white and ink becomes black.

That single step is most of the difference between "a photo of a page" and "a scan", and it is also why the threshold look works well: thresholding a washed-out photograph directly gives grey mush, while thresholding a stretched one gives clean black text on white.

The look is your choice: keep the colors, go grey with no color cast, clean up the lighting while keeping the tones, or drop to two tones. Two tones is what makes a page of text tiny, because the image stops being a gradient and becomes a pattern of black and white.

Pages, not pictures

The geometry is the same calculation the JPG to PDF page uses, so A4 means 210 by 297 millimeters here too, and each picture is scaled to fit inside the page without being distorted. "Follow the picture" chooses portrait or landscape per page; you can also force one orientation for the whole document, which is what a stack of mixed photographs usually needs.

When the paper size matches the picture instead, the pixels are converted to points at the resolution you pick - 96 dpi for screen-sized pixels, 72 dpi where one pixel becomes one point, or 150 to 300 dpi when the page has to survive a printer. The plan on the page prints the size you will get before you run anything.

Each page picture is written as a JPEG, which is the lossy step that keeps a document of photographs from being enormous. The quality setting is the dial: 80% is invisible on text and a fraction of the size of 92%.

What it will not do, and what to do instead

There is no deskew. A photograph taken at an angle stays at an angle, and the honest advice is to straighten it first - the image resizer on this site can turn a picture a quarter turn at a time, which handles the common "held the phone sideways" case. A few degrees of tilt needs a proper rotate-and-crop tool, and this page will not pretend otherwise.

There is no OCR. The output has no text layer: the words are pixels, so nothing in the PDF can be searched, selected or copied, and a screen reader sees a picture of a page. That is a real limitation for archives and for invoices, and it is the honest reason this tool is called "scan to PDF" rather than "make a text document".

There is no shadow removal beyond the tone stretch. A page lit from one side comes out better than it went in, but not flat the way a scanner makes it. For anything that has to be perfect - a passport copy, a contract - a real scanner or a scanning app with edge detection is the right tool.

A scan in practice: the plan, then the drop

The plan on this page is built from the settings alone, using the same pure planner the real job runs - so the paper size, the look and the quality shown in the plan are exactly what a job with those settings would produce, before any picture exists. When photographs arrive, they are stacked in the order they were added and each one becomes a page.

Every page is decoded and toned in the tab, so a short job runs in a second or two and a longer one is bounded by the device memory rather than by a server. The limits the plan prints - up to 20 pictures per job, 25 MB per file - are the tab owning the work, which is also why nothing ever has to leave the machine.