Working with pages

Everything under Pages is about the sheets themselves, not what is marked on them. It covers adding a sheet, taking out near-duplicates, cropping, reading the text on a scan, and writing the sheets back out as pictures. Turning, mirroring and resizing a sheet are on Changing the sheets.

Adding a blank sheet

Pages ▸ Insert Blank Page Before / After / at End.

It takes the size and rotation of the sheet beside it, not a paper size from Settings. A blank A4 sheet dropped into a deck of screenshots would be a mistake you might not notice, and the sheet beside it already has the right answer. Settings decides only when there is no neighbouring sheet to ask, which happens only in a document with no pages at all.

One undo step puts it back. Marks on later sheets move with the sheets: a mark on sheet 56 ends up on sheet 57. The number printed on its paper stays the same, because inserting a page does not renumber the book.

Adding a sheet is a change to the pages, so the document can no longer be saved by writing only its marks. That costs a little more on the next save and nothing else.

Near-duplicate pages

Photographing a discussion produces more frames than anyone needs. Most of the surplus is the same slide caught twice while someone was still talking.

Pages ▸ Find Near-Duplicate Pages marks the extra frames. They turn grey but stay selectable, so you can click one and look at it before deciding. Finding removes nothing by itself.

• Clear Duplicate Marks takes the offer back.
• Remove Marked Duplicates removes exactly what is marked, in one undo step.

The first page of each run is always kept: the one you saw when the slide first went up. Pages are compared across the whole deck, not only with their neighbours, so the same slide shown again later is offered as well.

You set how alike counts as the same. Once a verdict is offered, a slider appears under the pages list. Moving it updates the greyed pages immediately. Left is strict: 0 offers only pictures that are identical. Right is loose. The same number appears in Settings ▸ Reading ▸ Near-duplicate pages, out of 64.

Crop first. Screenshots of a meeting mostly show the meeting's frame: the toolbar, the faces, the participant list. That frame is the same in every shot, so whole screenshots of different slides can look alike and get offered together. Crop one page to the slide, use Copy Crop Rectangle, paste it onto a page, then pick every page in the sidebar and press ⌘K (Crop to Selection) to apply the rectangle to all of them. Then run the search again: it reads each page as cropped, so different slides come out different.

The comparison checks the shape of the light across each page: whether each patch is brighter than its neighbour. This is fast and survives a change in exposure. Two slides that differ by one number can still look alike to it. That is the right bias for a verdict you are asked to confirm: being offered a page you want to keep costs you only a glance, while missing a real duplicate costs you more.

Trimming the paper to what is drawn on it

Auto Crop trims every page to its own ink. A scanned paper carries an inch of white on every side. On a laptop, that inch is the difference between reading it and not.

Auto Crop always applies to the whole document, no matter what is picked in the sidebar. It is the only crop that works this way, and that is intentional, not an oversight. Cropping to a rectangle has to be told which sheets, because a rectangle drawn by hand has to be placed somewhere. What is drawn on a page, though, is a question each page answers for itself. Pages that already fit their content are skipped, and if none of them needed anything, the status line says so instead of reporting a change that did not happen.

Nothing is removed from the file. Auto Crop only moves each sheet's boundary inward, so Uncrop can give every page its full paper back. That is different from cropping to a region, where anything outside the rectangle is gone for good. See Select a region. The whole document counts as one undo step, not one per page.

Reading the text on all the pages

OCR recognises the words on any page that has none, so they can be searched, selected and copied. The page itself is not redrawn: it looks exactly as it did, and gains a layer of text underneath.

Pages that already have text are left alone. Running OCR on a bundle where four pages out of six hundred are scans costs only four pages of work. If every page already has text, the command says so, rather than appearing to do nothing.

Three things decide what it reads, and the order is not the one most people guess:

• A rectangle drawn with the region tool wins outright. If you have just marked out a figure and run this command, that is as plain a statement of intent as you can give, and reading nine hundred other pages is not it.
• Otherwise, the sheets you picked in the sidebar, and only the ones you actually clicked. A sheet the list scrolled to on its own, which happens when a document opens and again each time you turn a page, does not count. Counting those would quietly make the rule mean page one.
• Otherwise, all of them.

The document has to be a file first. Recognition writes the words it finds into the document's own pages, so a folder of pictures opened as pages has nowhere to put them. You are offered the save, not merely told about it, and told that saving writes all the pages, not only the ones picked. A save is the one act here where a selection deliberately does not mean "these." Anything unsaved is written before the reading begins, for the same reason.

It runs beside you rather than in front of you: a page takes about a second, the status line counts them off, and the window keeps working while it runs. OCR Again… is the same command, used on a page whose recognised text has since become wrong, most often one that has been turned. It replaces the old reading instead of adding a second one behind it. Coming back with nothing new found is a real answer: every page was read, and the words found were the ones already there.

Writing the pages out as pictures

File ▸ Save As Pictures… writes one PNG per page into a folder you choose. Choosing the folder the pictures came from replaces them. The command beside it, Save Pictures in Place, does something different: it writes over the pictures picked in the sidebar, or, if none are picked, only the ones that changed, without asking where. See Saving, printing and closing.

If any of the files are already there, you get one warning naming them before anything is written. Return cancels, so pressing a key without reading is the safe answer. A long list is cut at ten, with a count of the rest.

⌘S is not this. A folder of pictures opened here becomes a document, with pages, marks and a history, and saving it writes a PDF. Writing pictures back out is a separate action, and it is spelled differently on purpose.

Watch: a short video for this page is still to come.

Contents