Convert JPG Images to PDF

How an image becomes a page

A JPEG is already a compressed stream of bytes, and a PDF page can hold one directly. So the conversion does not re-encode the picture: the image data is embedded as an object in the new document, a page is created, and the page's content stream contains a single instruction to draw that object across the page area.

Because nothing is re-compressed, no quality is lost in the conversion itself, and the resulting PDF weighs roughly what the images weighed. Fifty photographs of four megabytes each produce a document in the region of 200 MB — which is a useful thing to know in advance, since it far exceeds what most mailboxes will carry.

The one decision the conversion has to make is how large the page should be. PDF measures pages in points at 72 to the inch, and an image measures itself in pixels, so a pixel count has to be turned into a physical size. That single conversion explains most of the surprises people meet with the output.

Pixels, points and whether it will print

Work the arithmetic once and the behaviour stops being mysterious. Divide an image's pixel width by the width of the page it lands on in inches, and you have the effective resolution of the printed page.

A phone photograph of roughly 3,000 by 4,000 pixels placed on a page eight inches wide gives around 375 pixels per inch — above the 300 dots per inch that print work assumes, so it prints cleanly. The same photograph blown up to fill a page twenty inches wide gives 150, which is visibly soft on paper.

Screens are far less demanding. A display renders somewhere between 72 and 96 pixels per inch, so an image that looks perfectly crisp on a laptop can still be well short of what a printer wants. Judging a document by how it looks on screen is the most common way to be surprised by the printed copy.

Screenshots are the reverse case: they are captured at screen resolution by definition, so they look exact on a display and soft on paper however they are placed.

A worked example: an expenses claim

Eighteen receipts photographed on a phone over the course of a month, to be submitted as one document. Eighteen is well inside the 50-image ceiling and each shot is a few megabytes, so the upload is straightforward.

Order is the thing to get right before processing rather than after. Images are placed in the order they appear in the upload list, and a phone names files by capture time, so a batch selected from a gallery usually arrives chronologically — which is generally what an expenses claim wants. Check the list on screen and rearrange it there; correcting the sequence afterwards means going through Organize PDF to move pages around.

The result is an eighteen-page PDF weighing roughly the sum of the photographs, which for full-resolution phone shots can easily be several tens of megabytes. Receipts do not need photographic quality, so put the finished document through Compress PDF before sending it — the common mailbox limit sits near 25 MB and a claim like this will often exceed it.

Photograph the page as if you were scanning it

The single largest influence on the finished document is how the picture was taken, and it costs nothing to take it well.

Hold the camera parallel to the page rather than at an angle. A page shot from above at a tilt arrives as a trapezium, and a trapezium cannot be squared up afterwards without stretching the text. Keep the whole sheet in frame with a small margin around it, so nothing is clipped and the page has an edge.

Light it evenly and from the side rather than head on. A phone held directly above a glossy page under a ceiling light casts its own shadow across the middle of the shot, and flash on paper produces a bright disc in the centre with dark corners.

Steady the phone against something. Camera shake is invisible at gallery size and obvious at full page.

All of this matters twice over if the document will later be made searchable with OCR PDF, because recognition accuracy is determined by the sharpness and evenness of the original far more than by anything applied afterwards.