How to split a PDF based on a size limit?

I have searched many places but unable to find a pretty good solution as such. So what I am trying to achieve is as below: My program will have quite a lot of PDF docs which I will have to send via mail. There is a mail server limitation of 4 MB. So if all the PDFs are less than 4 MB it will be sent as a single mail. Else I will have to create multiple files each less than 4 MB. Now my program works fine for the following cases: 1: Lots of files but each less than 4MB and hence keeping a tab during merging so that none of the merged files get over 4MB. 2: All files are pretty small and hence merging them together does not go to 4MB limit.

But there can be a scenario where there is one file which is, say, 14MB. I can split that document by pages. But that is also not a good solution as the pagesize is also not evenly distributed across the pages. I have used iText and PDFBox. Any help/pointer will be highly appreciated!

回答1:

Imagine a 3000 KB document with ten pages and the following objects:

four font subsets used on every page, each about 50 KB
ten images that figure on a single page, each about 200 KB (one image per page)
four images that figure on every page, each about 50 KB
ten pages with content streams of about 25 KB each
about 350 KB for objects such as the catalog, the info dictionary, the page tree, the cross-reference table, etc...

A single page will need at least: - the four font subsets: 4 times 50 KB - the single image: 1 time 200 KB - the four images: 4 times 50 KB - a single content stream: 1 time 50 KB - a slightly reduced cross-reference table, a slightly reduced page tree, an almost identical catalog, an info dictionary of identical size,... 200 KB

Together that's 850 KB. This means that you end up with 8500 KB (10 times 850 KB) if you split up a 10-page 3000 KB PDF document into 10 separate pages.

This example is the result of guess work (based on experience) and it assumes that the PDF is predictable. Most PDFs aren't:

some pages will require high-definition images (maybe even megaBytes), other pages won't have any images,
some pages will need many different fonts and font subsets (lots of kiloBytes), other pages will consist of merely some vector drawings (tiny content stream if compressed).
different pages can share a large amount of resources (Form XObjects, Image XObjects,...), other pages won't share any resources.
and so on...

You have noticed that yourself, as you write: I can split that document by pages. But that is also not a good solution as the pagesize is also not evenly distributed across the pages.

That's exactly why your question can have no other answer than: you'll have to do trial and error. No software can predict how much space is needed by a page before you look at what is needed by that page.

Update:

As David indicates in the comments, it is possible to calculate all the resources needed for a page, and to check if the current resources plus the needed resources exceed the maximum file size.

I have written a small example:

public void manipulatePdf(String src, String dest)
    throws IOException, DocumentException {
    Document document = new Document();
    PdfCopy copy = new PdfSmartCopy(document, new FileOutputStream(dest));
    document.open();
    PdfReader reader = new PdfReader(src);
    for (int i = 1; i <= reader.getNumberOfPages(); i++) {
        // check resources needed for reader.getPageN(i);
        copy.addPage(copy.getImportedPage(reader, i));
        System.out.println("After adding page: " + copy.getOs().getCounter());
    }
    document.close();
    System.out.println("After closing document: " + copy.getOs().getCounter());
    reader.close();
}

I have executed the example on a PDF sample with 18 pages and this was the output:

After adding page: 56165
After adding page: 111398
After adding page: 162691
After adding page: 210035
After adding page: 253419
After adding page: 273429
After adding page: 330696
After adding page: 351564
After adding page: 400351
After adding page: 456545
After adding page: 495321
After adding page: 523640
After adding page: 576468
After adding page: 633525
After adding page: 751504
After adding page: 907490
After adding page: 957164
After adding page: 999140
After closing document: 1002509

You see how the file size of the copy gradually grows with each page that is added. After all pages are added, the size is 999140 bytes, and then the page tree and cross-reference stream are written, adding another 3369 bytes.

Where it says // check resources needed for reader.getPageN(i);, you could make a guesstimate of the size that will be added for the page and break out of the loop if it exceeds a maximum value.

Why would this be a guesstimate:

You could be counting objects that are already added. If you keep track of the objects (not that difficult), your guess will be more accurate.
I'm using PdfSmartCopy. Suppose that there are two identical objects inside your PDF. Bad PDF software often causes such problems. For instance: the same image bytes are added twice to the file. PdfSmartCopy can detect this and will reuse the first object it encounters instead of adding the redundant bytes of the extra object.

We currently don't have a reader.getTotalPageBytes() in PdfReader because PdfReader tries to use as little memory as possible. It won't load any objects into memory as long as these objects aren't needed. Hence it doesn't know the size of each object before the page is imported.

However, I'll make sure that such a method is added in the next release.

Update:

In the next version, you'll find a tool named SmartPdfSplitter that depends on a new class named PdfResourceCounter. You can use it like this:

PdfReader reader = new PdfReader(src);
SmartPdfSplitter splitter = new SmartPdfSplitter(reader);
int part = 1;
while (splitter.hasMorePages()) {
    splitter.split(new FileOutputStream("results/merge/part_" + part + ".pdf"), 200000);
    part++;
}
reader.close();

Note that this can result in a single-page PDF that exceeds the limit (which was set to 200000 bytes in the code sample) in case that single page can not be reduced to less bytes. In that case, splitter.isOverSized() will return true and you'll have to find another way to reduce the PDF.

回答2:

PDF Clown supports page data size prediction without need of trial and error: since 2010 it has been featuring a dedicated method (org.pdfclown.tools.PageManager.getSize(Page)) that calculates in memory the actual page data size without the need to write it to a file for trial.

Furthermore, there's another method (org.pdfclown.tools.PageManager.split(long maxDataSize)) purposely implemented to address your kind of scenario which leverages the above-mentioned PageManager.getSize method: it automatically splits a file based on a size limit without creating any intermediate, ugly, stupid, temporary file for trial and error.

You can see a practical example of its use in the org.pdfclown.samples.cli.PageManagementSample (PageDataSizeCalculation and DocumentSplitOnMaximumFileSize cases) included in the downloadable distribution -- here it is an example of console output from the PageDataSizeCalculation case:

Page 1: 29380 (full); 29380 (differential); 29380 (incremental)
Page 2: 30493 (full); 1501 (differential); 30881 (incremental)
Page 3: 21888 (full); 1432 (differential); 32313 (incremental)
Page 4: 33781 (full); 4789 (differential); 37102 (incremental)
. . .

where:

full is the page data size encompassing all its dependencies (like shared resources) -- this is the size of the page when extracted as a single-page document;
differential is the additional page data size -- this is the extra content that's not shared with previous pages;
incremental is the data size of the page sublist encompassing all the previous pages and the current one.