Compression is why your scans look worse than the paper
Between the sensor reading the page and a file appearing on disk there is a step nobody chooses deliberately: the image gets compressed. Most defaults compress in a way that discards information, and the kind of information they discard — fine, high-contrast detail on a plain background — is exactly what the small print on a receipt is made of. It is the most common reason a scan looks noticeably worse than the paper it came from.
Lossy and lossless, and why the default is lossy
A lossless method reconstructs the original samples exactly; a lossy one approximates them. Lossy compression works by deciding that some differences do not matter perceptually and spending fewer bits on them. For a photograph of a landscape this is an excellent bargain: the file gets much smaller and you cannot see what went.
Documents break the assumption. The methods in common use divide the image into small blocks and describe each block in terms of how much fine variation it contains. A block that is mostly flat paper with one sharp character edge crossing it is precisely the case that gets described crudely, and the result is the familiar mess around small text: faint ripples radiating from the strokes, grey haze in the white space, and edges that were crisp on the paper arriving soft.
Scanner defaults lean lossy because the alternative produces larger files, and file size is what most people notice and complain about. The damage is invisible until you zoom in, and by then the paper may be gone.
The symptoms, and how to tell them from other faults
Compression damage looks distinctive once you know it.
- It clusters around detail. Flat areas of paper stay clean; the artefacts appear as a halo or ripple immediately around characters and lines.
- It is uniform across the page. Unlike a roller mark or a dirty sensor strip, it does not favour one column or one edge. If a defect runs down the page in the same place on every sheet, it is a hardware problem, not compression — see diagnosing scan defects.
- It gets worse where print is smallest. The header in large type looks fine; the line items look mushy.
- It survives rescanning at the same settings and disappears at different ones. Which is the test.
The reason to distinguish it from a dirty sensor or a fading slip is that the fix is completely different. Cleaning the glass will not help, and neither will a better exposure.
The trap of aggressive document compression
Some devices offer very high compression modes specifically for documents, which work by a cleverer route: they find repeated shapes across the page, store each shape once, and reference it wherever it recurs. On a page of consistent printed type this is remarkably efficient.
It also introduces a failure mode worth knowing about, because it is not a blur. If two shapes are judged similar enough, the same stored glyph gets used for both — which means a character can be substituted with a different character that looked alike at the compression stage. On worn or low-contrast till print, where an 8 and a 3 may differ by a few pixels, that is a substitution in a number. It does not look like damage. It looks like a clean, confident digit that is wrong.
For anything containing amounts, this argues for avoiding the most aggressive document modes regardless of how good the file sizes are. A slightly blurry total that you can tell is blurry is safer than a crisp total you have no reason to doubt.
Settings that avoid the problem
There are three ways out, in rough order of how often they apply.
Reduce the compression rather than the resolution. If files are too big, most people turn the dpi down. Moving to a gentler compression setting at the same resolution usually gives a better-looking result for the same size, because you are removing redundancy instead of removing samples. The resolution question is separate and worth settling on its own terms: what dpi does to a receipt scan.
Use a lossless path for anything that matters. Most devices can write a lossless output if you ask, and for a small number of irreplaceable items the size difference is irrelevant.
Consider what the colour mode implies. Pure black-and-white output is compressed by methods that are lossless by nature, since there are only two values, so it has none of these artefacts. That is a genuine advantage — it is just outweighed for receipts by everything a hard threshold throws away, as colour, greyscale, or bitonal sets out.
Never compress twice
Each pass compounds the previous one. A scan that was compressed by the device, opened, and re-saved by something else has been through two rounds of approximation, the second working on the artefacts of the first. Anything that regenerates a file — an app that re-encodes on import, a service that resizes for upload, a workflow that converts between formats — is a second pass whether it advertises itself as one or not.
The defence is to keep the file as the device wrote it and treat everything else as a derivative. If a tool needs a smaller version, let it make one and keep the original untouched. This is also the reason to check what happens to your scans when they enter software that stores them in its own library, which is discussed in bundled software lock-in.
Test it once on your own paper
Scan the same faint, small-print receipt at your normal settings and at the gentlest compression the device offers. View both at high magnification and compare the line items, not the header. If you can see the difference, your default is costing you something; if you cannot, it is not, and you can stop thinking about it. This is a five-minute check and it belongs in the same session as the other baseline measurements in testing a scanner with your own paper.
Whether a compressed scan remains an acceptable record is a separate matter entirely, and it is not a technical judgement. Rules about the form records must take vary by country and by the kind of entity keeping them, so put the question to your own tax authority or an accountant where you are — and until you have an answer, prefer the file that lost less.