I am using convert (Imagemagick component, delegating to Ghostscript in background) to transform the first page of PDF files to images.

Usually, convert -density 200 file.pdf[0] first_page.png will do the job, and it will sample the PDF file at 200 pixels per inch of paper.

However it seldom happens that some PDF are abnormally huge (sometimes A0 paper, and recently a PDF with a page exceeding 23 m² (183 inch in length, 185 in width).

For such files, convert will hang, eat CPU time. Images of 35000+ pixels in width and height are simply not usable.

Therefore the question: is there a switch in Imagemagick that would adapt the density to the page size, or at least specify that we don't want to sample more than a portion of maximal area of the PDF file (top left corner, 30x30 inch for example)?


EDIT: On its official git repository, MuPDF has added the -w and -h switches that, jointly with -r will do what is wanted here.


I modified mupdf's pdfdraw to support drawing in best fit mode, so I could state that the output needed to be 128x128 at most and it would fit the output in the box while maintaining the aspect ratio. Before I did that the only way was to use pdfinfo to get the page size and then do the calcuations to fit it in a box and then ask pdfdraw to draw it with that scale factor (dots per inch).

Well, after that long story the process to do that is rather simple:

  1. get the page size of the page to render (in pdf terms the media box) this can be done via pdfinfo and grep and will appear in pts (points, 1/72th of an inch) or via a pdf library like pyPDF like:

    import pyPdf
    p = pyPdf.PdfFileReader(file("/home/dan/Desktop/Sieve-JFP.pdf", "rb"))
    x,y,w,h = p.pages[0]['/MediaBox']
  2. for a box fit do dpi = min( A/(w/72.), B/(h/72.) )
    where A is the maximum width and B is the maximum height; w and h are the width and height of the page.

  3. pass dpi to convert -density $dpi

and as requested a slightly fudged git commit diff:

commit 0000000000000000000000000000000000000000
Author: Dan D.
Date:   Thu Jul 28 16:33:33 2011 -0400

    add options to pdfdraw to limit the output's width and height

    note that scaling must occur before rotation

diff --git a/apps/pdfdraw.c b/apps/pdfdraw.c
index 0000000..1234567 100644
--- a/apps/pdfdraw.c
+++ b/apps/pdfdraw.c
@@ -12,8 +12,10 @@

 char *output = NULL;
-float resolution = 72;
+float resolution = -1;
 float rotation = 0;
+float width = -1;
+float height = -1;

 int showxml = 0;
 int showtext = 0;
@@ -47,6 +49,8 @@ static void usage(void)
        "\t\tsupported formats: pgm, ppm, pam, png, pbm\n"
        "\t-p -\tpassword\n"
        "\t-r -\tresolution in dpi (default: 72)\n"
+       "\t-w -\tmaximum width (default: no limit)\n"
+       "\t-h -\tmaximum height (default: no limit)\n"
        "\t-A\tdisable accelerated functions\n"
        "\t-a\tsave alpha channel (only pam and png)\n"
        "\t-b -\tnumber of bits of antialiasing (0 to 8)\n"
@@ -150,13 +154,39 @@ static void drawpage(pdf_xref *xref, int pagenum)

    if (output || showmd5 || showtime)
-       float zoom;
+       float zoom = 1.0;
        fz_matrix ctm;
        fz_bbox bbox;
        fz_pixmap *pix;
+       float W, H;

-       zoom = resolution / 72;
-       ctm = fz_translate(0, -page->mediabox.y1);
+       ctm = fz_identity;
+       ctm = fz_concat(ctm, fz_translate(0, -page->mediabox.y1));
+       ctm = fz_concat(ctm, fz_rotate(page->rotate));
+       ctm = fz_concat(ctm, fz_rotate(rotation));
+       bbox = fz_round_rect(fz_transform_rect(ctm, page->mediabox));
+       W = bbox.x1 - bbox.x0; 
+       H = bbox.y1 - bbox.y0;
+       if (resolution != -1)
+           zoom = resolution / 72;
+       if (width != -1) 
+       {
+           if (resolution != -1)
+               zoom = MIN(zoom, width/W);
+           else
+               zoom = width/W;
+       }
+       if (height != -1)
+       {
+           if (resolution != -1 || width != -1)
+               zoom = MIN(zoom, height/H);
+           else
+               zoom = height/H;
+       }
+       ctm = fz_identity;
+       ctm = fz_concat(ctm, fz_translate(0, -page->mediabox.y1));
        ctm = fz_concat(ctm, fz_scale(zoom, -zoom));
        ctm = fz_concat(ctm, fz_rotate(page->rotate));
        ctm = fz_concat(ctm, fz_rotate(rotation));
@@ -295,7 +325,7 @@ int main(int argc, char **argv)
    fz_error error;
    int c;

-   while ((c = fz_getopt(argc, argv, "o:p:r:R:Aab:dgmtx5")) != -1)
+   while ((c = fz_getopt(argc, argv, "o:p:r:R:w:h:Aab:dgmtx5")) != -1)
        switch (c)
@@ -303,6 +333,8 @@ int main(int argc, char **argv)
        case 'p': password = fz_optarg; break;
        case 'r': resolution = atof(fz_optarg); break;
        case 'R': rotation = atof(fz_optarg); break;
+       case 'w': width = atof(fz_optarg); break;
+       case 'h': height = atof(fz_optarg); break;
        case 'A': accelerate = 0; break;
        case 'a': savealpha = 1; break;
        case 'b': alphabits = atoi(fz_optarg); break;
@@ -321,6 +353,10 @@ int main(int argc, char **argv)
    if (fz_optind == argc)

+   if (width+height == -2)
+       if (resolution == -1)
+           resolution = 72;
    if (!showtext && !showxml && !showtime && !showmd5 && !output)
        printf("nothing to do\n");
  • Thank you! When you say you modified pdfdraw… where could I find your modification? – Benoit Feb 13 '12 at 12:18
  • You're Welcome, well, note that they apply to mupdf's pdfdraw and not to xpdf's pdfdraw which is the command which is most commonly refered to as pdfdraw. They reside in my ~/git.l/mupdf git repos which is not online. Mainly because my changes apply to an older version of mupdf (I think 0.8; I was going to update for 0.9 but the commits in the mupdf repos are so messy and although the commiters know how to code they don't know how to write git commits) and I am confused about what I'm trying to do. – Dan D. Feb 13 '12 at 12:42
  • is there a way you post a diff anywhere then? Thanks! – Benoit Feb 13 '12 at 12:53
  • Very nice, thank you. Will try to adapt it! I am considering adding a bounty for that answer when I am able to open it. If I forget please remind me. – Benoit Feb 13 '12 at 13:08
  • Submitted patch proposal 692855 to Ghostscript. Thanks again. – Benoit Feb 13 '12 at 14:37

You are using the wrong command. Use -resample instead. It is also advisable to provide a specific width and height if possible.

-density is merely a flag. -resample actually changes pixel dimensions: the only measurement that matters.

edit: doc for -resample http://www.imagemagick.org/script/command-line-options.php#resample

The -density option sets an attribute and does not alter the underlying raster image. It may be used to adjust the rendered size for desktop publishing purposes by adjusting the scale applied to the pixels. To resize the image so that it is the same size at a different resolution, use the -resample option.

At a simplistic level in CG, inches do not exist. For raster images, only the pixels are stored. The dpi is merely a suggestion.

Say you have 3 squares on a table, and 300 pennies. If I have a density of 300 pennies per square, there is only one square with 300 pennies in it.

If I change density to 100 pps, I now have 3 squares, but still 300 pennies total (100 pennies in each square). You have not changed the number of pennies, only the manner in which you distribute the pennies across an arbitrary unit of measure.

If I resample the original to 100pps I have 1 square, and 100 pennies total. I have changed the number of pennies.

I suspect that in cases where the page size goes up, you are dealing with something that had a high resolution such as 1200dpi line art, and by changing the density to 300 quadruples the inch measurement when you open the result with something that honors the flag.

  • I am sorry, I don't understand that answer; could you be more precise and give command line? I recall that I do not know the page dimensions in advance, and I want to make rasters of approximately 200px per inch of paper, unless the output raster would be too large, then downscale. What command would you use for that please? thank you. – Benoit Feb 23 '12 at 15:58
  • see edit for link to convert documentation – horatio Feb 23 '12 at 16:19
  • okay, but how does that relate to my problem? PDF format has no notion of pixels when you read it with Imagemagick. When you load it, the -density parameter is used to sample it. – Benoit Feb 23 '12 at 16:27
  • PDF is a container format which can contain images and also contains DPI information for the purposes of rendering a document at the specified paper size. The documentation for convert explains what commands and parameters to use. And as I said: if you specify the desired width and or height (aka size), along with the -resample, you explicitly tell convert what you want. -resample command is used to "Resize the image so that its rendered size remains the same as the original at the specified target resolution." Whenever inches are used as a dimension, it is the rendered size – horatio Feb 23 '12 at 17:09

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.