Building a document from photographs

Photographing paperwork and turning it into one file is now the commonest way documents get digitised. It works well, provided you avoid a few predictable traps.

Take the photographs properly

Directly above the page, not leaning over it. Leaning produces a trapezoid that no amount of processing fully squares up.

Diffuse light from in front. Your own shadow across the page is the most common ruin, followed by a single side lamp producing a gradient across the paper.

Fill the frame with the page but leave a small margin, so nothing is clipped and the edges are available for straightening.

Check the order before building

Photographs sort by the time they were taken, which is usually right, and occasionally very wrong if you went back to redo page four. Rename or reorder before assembling rather than discovering the problem in the finished PDF.

Choose a page size deliberately

Matching each page to its image sounds sensible and produces a document where every page is a slightly different size, because no two photographs are framed identically. On screen this looks fine. Printed, it is a mess.

Pick a standard size, A4 or Letter, and let the images fit inside it. The result is consistent and prints predictably.

Fit, do not fill

"Fill the page" crops whatever does not fit the aspect ratio, which on a document means losing an edge. "Fit inside the margins" keeps the whole page and leaves a border. For paperwork, always fit.

Compress afterwards, not before

A photograph from a modern phone is several megabytes and around four thousand pixels wide. Twenty of them make a very large PDF.

Build the document first, then compress it. This is the case compression is genuinely good at: photographic content, at far higher resolution than anyone will view it. Expect substantial reductions with no visible difference.

Add a text layer if it will need finding later

A PDF of photographs is a PDF of pictures, and searching it finds nothing. If the document is going into any kind of filing system, or you may need to find a reference number in it a year from now, run recognition over it.

This is the difference between an archive you can use and a pile of images you have to open one at a time.

Before you send it

Scroll through and check for the page you photographed twice and the one you missed. It happens on almost every multi-page capture, and it is much easier to notice now than after the recipient does.

Getting the capture right

Directly above the page. Leaning over it produces a trapezoid that no processing squares up, because the perspective distortion is not a rotation.

Diffuse light in front of you. The two ruinous lighting mistakes are your own shadow falling across the page, and a single strong lamp to one side, which produces a brightness gradient the compressor then encodes as detail.

Fill the frame with the page but leave a small border, so the edges survive for straightening and cropping.

If your phone has a document mode, use it. It handles the perspective correction and thresholding that would otherwise take three separate steps afterwards.

Assembling in the right order

Check the order before building. Photographs sort by capture time, which is usually right and occasionally very wrong if you went back to redo a page.

Choose a standard page size rather than matching each image. Matching each image sounds sensible and produces a document where every page is a slightly different shape, because no two photographs are framed identically. It looks acceptable on screen and prints badly.

Fit inside the margins rather than filling the page. Filling crops the overflow, which on a document means losing an edge.

Processing afterwards

In order: straighten, crop, remove any blank captures, run recognition, then compress.

Straightening first because recognition reads crooked text badly. Cropping second because it reduces what everything downstream has to process. Recognition before compression, because compressing first means recognising a degraded image.

Compression last is where the size comes down. This is exactly the case it is good at: photographic content at far higher resolution than anyone will view, on pages that are already images.

Whether to add a text layer

If the document is going into any filing system, or you might need to find something in it later, run recognition. The difference between an archive you can search and a pile of images you open one at a time is enormous, and it costs one step.

If you only need to send it once and it will never be referred to again, skip it and save the file size.

Before sending

Scroll through and count. The two errors that happen on almost every multi-page capture are the page photographed twice and the page missed entirely, and both are far easier to fix while the paper is still on the desk.