How can I tell the resolution of scanned PDF from within a shell script?

Question

I have a large collection of documents scanned into PDF format, and I wish to write a shell script that will convert each document to DjVu format. Some documents were scanned at 200dpi, some at 300dpi, and some at 600dpi. Since DjVu is a pixel-based format, I want to be sure I use the same resolution in the target DjVu file as was used for the scan.

Does anyone know what program I can run, or how I can write a program, to determine what resolution was used to produce a scanned PDF? (Number of pixels might work too as almost all documents are 8.5 by 11 inches.)

---

Clarification after responses: I'm aware of the difficulties highlighted by Breton, and I'm willing to concede that the problem in general is ill-posed, but I'm not asking about general PDF documents. My particular documents came out of a scanner. They contain one scanned image per page, same resolution each page. If I convert the PDF to PostScript I can poke around by hand and find pixel dimensions easily; I could probably find image sizes with more work. And if in desperate need I could modify the dictionary stack that gs is using; long ago, I wrote an interpreter for PostScript Level 1.

All of that is what I'm trying to avoid.

---

Thanks to help received, I've posted an answer below:

1. Extract the bounding box from the PDF using identify, taking only the output for the first page, and understanding that the units will be PostScript points, of which there are 72 to an inch.
2. Extract images from the first page using pdfimages.
3. Get height and width of image. This time identify will give number of pixels.
4. Add the total areas of the images to get the number of dots squared.
5. To get resolution, compute areas of bounding box in inches squared, divide dots squared by inches squared, take the square root, and round to the nearest multiple of 10.

Full answer with script is below. I'm using it in live fire and it works great. Thanks Harlequin for pdfimages and Spiffeah for the alert about multiple images per page (it's rare, but I've found some).

Answer 1

If a pdf has been created by scanning then there should only be one image associated with each page. You can find each image resolution for each page image by parsing the pdf using the iText(Java) or iTextSharp(the .net port) libraries easily.

If you want to roll your own utility to do this, do something like the following in iTextSharp :

PdfReader reader = new PdfReader(filename);
for (int i = 1; i <= reader.NumberOfPages; i++)
{
PdfDictionary pg = reader.GetPageN(i);
PdfDictionary res = (PdfDictionary)PdfReader.GetPdfObject(pg.Get(PdfName.RESOURCES));
PdfDictionary xobjs = (PdfDictionary)PdfReader.GetPdfObject(res.Get(PdfName.XOBJECT));
if (xobjs != null) 
{
    foreach (PdfName xObjectKey in xobjs.Keys)
    {
    PdfObject xobj = xobjs.Get(xObjectKey);
    PdfDictionary tg = (PdfDictionary)PdfReader.GetPdfObject(xobj);
    PdfName subtype = (PdfName)PdfReader.GetPdfObject(tg.Get(PdfName.SUBTYPE));
    if  (subtype.Equals(PdfName.IMAGE))
    {
        PdfNumber width = (PdfNumber)tg.Get(PdfName.WIDTH);
        PdfNumber height = (PdfNumber)tg.Get(PdfName.HEIGHT);
        MessageBox.Show("image on page [" + i + "] resolution=[" + width +"x" + height + "]");
    }
    }
}
}   
reader.Close();

Here for each page we read through each XObject of subtype Image and get the WIDTH and HEIGHT values. This will be the pixel resolution of the image that the scanner has embedded in the pdf.

Note that the scaling of this image to match the page resolution (as in the size of the page rendered in Acrobat - A4, Letter, etc) is performed separately in the page content stream, which is represented as a subset of postscript, and much harder to find without parsing the postscript.

Be aware that there are some scanners that will embed the scanned image as a grid of smaller images (for some kind of size optimization I assume). So if you see something like 50 small images popping up for each page, that could be why.

Hope this helps in some way if you have to roll your own utility.

Answer 2

pdfimages has a -list option that gives the height width in pixels and also y-ppi and x-ppi.

 pdfimages -list tmp.pdf           
page   num  type   width height color comp bpc  enc interp  object ID x-ppi y-ppi size ratio
--------------------------------------------------------------------------------------------
   1     0 image    3300  2550  gray    1   1  ccitt  no       477  0   389   232  172K  17%
   2     1 image    3300  2550  gray    1   1  ccitt  no         3  0   389   232  103K  10%
   3     2 image    3300  2550  gray    1   1  ccitt  no         7  0   389   232  236K  23%
   4     3 image    3300  2550  gray    1   1  ccitt  no        11  0   389   232  210K  20%
   5     4 image    3300  2550  gray    1   1  ccitt  no        15  0   389   232  250K  24%
   6     5 image    3300  2550  gray    1   1  ccitt  no        19  0   389   232  199K  19%
   7     6 image    3300  2550  gray    1   1  ccitt  no        23  0   389   232  503K  49%
   8     7 image    3300  2550  gray    1   1  ccitt  no        27  0   389   232  154K  15%
   9     8 image    3300  2550  gray    1   1  ccitt  no        31  0   389   232 21.5K 2.1%
  10     9 image    3300  2550  gray    1   1  ccitt  no        35  0   389   232  286K  28%
  11    10 image    3300  2550  gray    1   1  ccitt  no        39  0   389   232 46.8K 4.6%
  12    11 image    3300  2550  gray    1   1  ccitt  no        43  0   389   232 55.5K 5.4%
  13    12 image    3300  2550  gray    1   1  ccitt  no        47  0   389   232 35.0K 3.4%
  14    13 image    3300  2550  gray    1   1  ccitt  no        51  0   389   232 26.9K 2.6%
  15    14 image    3300  2550  gray    1   1  ccitt  no        55  0   389   232 66.5K 6.5%
  16    15 image    3300  2550  gray    1   1  ccitt  no        59  0   389   232 73.9K 7.2%
  17    16 image    3300  2550  gray    1   1  ccitt  no        63  0   389   232 47.0K 4.6%
  18    17 image    3300  2550  gray    1   1  ccitt  no        67  0   389   232 30.1K 2.9%
  19    18 image    3300  2550  gray    1   1  ccitt  no        71  0   389   232 70.3K 6.8%
  20    19 image    3300  2550  gray    1   1  ccitt  no        75  0   389   232 46.0K 4.5%
  21    20 image    3300  2550  gray    1   1  ccitt  no        79  0   389   232 28.9K 2.8%
  22    21 image    3300  2550  gray    1   1  ccitt  no        83  0   389   232 72.7K 7.1%
  23    22 image    3300  2550  gray    1   1  ccitt  no        87  0   389   232 47.5K 4.6%
  24    23 image    3300  2550  gray    1   1  ccitt  no        91  0   389   232 30.1K 2.9%

Answer 3

I guess that the scans are included as images in the PDF, so you could use pdfimages to extract them first. Then, identify should be able to find the correct data.

Answer 4

Too long to put into a comment, but neither ImageMagick nor GraphicsMagic is up to the job; every answer is wrong:

: nr@yorkie 1932 ; gm identify -format "x=%x y=%y w=%w h=%h" drh*rec*pdf
x=0 y=0 w=612 h=792
x=0 y=0 w=612 h=792
x=0 y=0 w=612 h=792
x=0 y=0 w=612 h=792
x=0 y=0 w=612 h=792
x=0 y=0 w=612 h=792
x=0 y=0 w=612 h=792
x=0 y=0 w=612 h=792

: nr@yorkie 1933 ; identify -format "x=%x y=%y w=%w h=%h" drh*rec*pdf   
x=72 Undefined y=72 Undefined w=612 h=792x=72 Undefined y=72 Undefined     w=612 h=792x=72 Undefined y=72 Undefined w=612 h=792x=72 Undefined     y=72 Undefined w=612 h=792x=72 Undefined y=72 Undefined w=612     h=792x=72 Undefined y=72 Undefined w=612 h=792x=72 Undefined y=72     Undefined w=612 h=792x=72 Undefined y=72 Undefined w=612 h=792
: nr@yorkie 1934 ; 

The correct parameters for this document is that each scanned page is 5100 pixels wide and 6600 pixels high, unsurprising for this was an 8.5-by-11 scanned at 600dpi. The output from ImageMagic is astoundingly unprofessional.

No downvotes because you were trying to be helpful, but *Magick don't work.