To rotate and clean up scanned client documents, work in a fixed order: turn sideways and upside-down pages upright, deskew crooked ones, remove blank backs only after checking them, then compress without losing small print. Do it before sorting, so every page reads cleanly when it is split, classified and matched.
What goes wrong in scanned client documents
Scanned client documents go wrong in five predictable ways: pages turned sideways or upside down, pages scanned at a slant, blank backs from double-sided scanning, dark borders and specks, and files far larger than they need to be. Each one slows the person or the software reading the page next.
Sideways pages are the most common. A client feeds a stack of mixed paper into a home scanner, and every landscape statement comes out rotated. A reader has to turn their head or the page for each one. OCR software may not fix it either: the Tesseract command-line documentation describes its default page mode as "Fully automatic page segmentation, but no OSD", meaning no orientation detection.
Skew is the quiet problem. A page fed at a two-degree angle still looks readable to a person, but the lines of a bank statement no longer run straight across. Columns drift, and an amount can end up read against the wrong date.
Blank backs come from duplex scanning. A 20-page stack of single-sided forms becomes a 40-page PDF where every other page is empty. That doubles the pages to review and, on per-page tools, the pages to process.
File size causes the last problem. A colour scan at high resolution of a 60-page packet can be too large to email, and slow to open on a laptop. The question of whether a scan is good enough to read at all, as opposed to merely large, is covered in scanned versus digital bank statement PDFs.
How do you rotate and deskew scanned client pages in bulk?
Scanned client pages can be rotated and deskewed in bulk with a tool that detects page orientation from the text itself. The open-source OCRmyPDF tool does both in one pass. For a handful of pages, a PDF viewer's rotate command on a group of selected thumbnails is faster than any setup.
For a few pages, select and rotate by hand. In Preview on a Mac, Apple's Preview guide says to choose View > Thumbnails or View > Contact Sheet, select the pages, then click the Rotate Left tool, or hold Option and click to rotate right. Selecting all the sideways pages at once turns them together.
For many files, automate it. The OCRmyPDF cookbook says --rotate-pages "attempts to determine the correct orientation for each page and rotates the page if necessary", and --deskew "will correct pages that were scanned at a skewed angle by rotating them back into place." It also notes that --rotate-pages-threshold makes rotation more or less aggressive, and that the default is "quite conservative."
Leave the threshold at its default on a first run. A conservative setting may miss a page with very little text, such as a mostly blank cover sheet, but it rarely turns an upright page sideways. Check the pages it skipped by eye.
The same cookbook describes --clean, which uses the unpaper tool to clean pages before OCR without changing the output, and --clean-final, which puts the cleaned page into the final file. Use --clean first. Because --clean-final replaces the original page image in the output, switch to it only after checking a sample of cleaned pages for lost marks.
Removing blank pages from scanned client documents safely
Removing blank pages from scanned client documents is safe only when each candidate is checked before it is deleted. A page that looks blank at thumbnail size can hold a single handwritten initial, a stamp or the last line of a statement. Flag nearly empty pages first, then delete only those that are truly empty.
If the scanner or PDF tool in use offers automatic blank-page removal, look for its sensitivity or threshold setting. Start at the lowest setting that catches the obvious empty backs, and review anything the tool flags before it is discarded. Raise it only if many empty pages get through.
Watch for three kinds of nearly blank page. The back of a form copy may carry printed instructions in light grey. A signature page may hold one signature and a date. The last page of a statement may hold only a closing balance. Each looks empty at a glance and matters to the file.
For the pages you do remove, a PDF viewer is enough. Apple's guide to adding, deleting or moving PDF pages says to choose View > Thumbnails or View > Contact Sheet, select the pages, and press Delete or choose Edit > Delete.
Once the blank backs are gone, a long packet usually still holds several documents. Separating them is the next step; see how to split a scanned PDF into separate documents by hand.
Should you compress scanned client documents before filing?
Scanned client documents can be compressed before filing as long as the smallest print stays readable. Lossless compression shrinks a file without changing any pixel, so it is always safe. Lossy compression and lower resolution shrink it further, and should stop at the point where every digit is still clear.
The OCRmyPDF optimizer documentation sets out the trade-off as levels. -O1, the default, "enables lossless optimizations, such as transcoding images to more efficient formats." -O2 adds "lossy optimizations and color quantization." -O3 "targets lower image quality." It also says monochrome images can be converted to JBIG2 at -O1 if a JBIG2 encoder is installed.
The test for how small is too small is legibility. For records kept electronically, IRS Revenue Procedure 97-22 (section 4.01(3)) says reproduced records must show "a high degree of legibility and readability", and defines legibility as being able to "identify all letters and numerals positively and quickly." It is a useful yardstick for a scan: zoom to 100 percent on the smallest box of a tax form and check that every digit is unambiguous.
Colour matters more than people expect. A greyscale or black-and-white scan of a typed statement is much smaller than colour and loses nothing. Keep colour for documents where it carries meaning, such as a highlighted receipt total or a coloured stamp.
A five-step cleanup checklist
Use this order on each client packet. The settings are starting points to adjust after checking a sample, not fixed rules.
| Step | What it fixes | Setting to start with |
|---|---|---|
| 1. Rotation | Sideways and upside-down pages | Automatic orientation on, default (conservative) threshold; fix skipped pages by hand |
| 2. Deskew | Pages scanned at an angle | Deskew on for every page |
| 3. Blank pages | Empty backs from duplex scans | Lowest sensitivity; review every flagged page before deleting |
| 4. Resolution | Oversized images | Keep the scan's original resolution; do not downsample until step 5 shows smallest print still clear |
| 5. Compression | Large files | Lossless first (for example OCRmyPDF -O1); try lossy only if the file is still too large, then recheck at 100 percent zoom |
After step 5, open three random pages and read the smallest number on each. If any digit is unclear, go back one step.
Cleanup is one stage in a longer chain that runs from intake to export. The post on accounting document automation shows which stages are worth automating and which stay manual, and how to sort phone photos of receipts covers the camera images that arrive alongside scans. Schema CPA reads each page after cleanup and classifies it by type, issuer and period; the document categories it reads are listed on the homepage.