Quality degradation of a text pdf after pdf>png>pdf - pdf

Question

I have a very specific requirement where i must automatically stamp every page of a PDF file (for a faxing application), so here's the process i've made:
step 1: Convert PDF to PNG, one png file per page
cmd1: gs -dSAFER -dBATCH -dNOPAUSE -sDEVICE=png16m -dGraphicsAlphaBits=4 -dTextAlphaBits=4 -r400 -sOutputFile=image_raw.png input.pdf
cmd2: mogrify -resize 31.245% image_raw.png
input.pdf (input): https://www.dropbox.com/s/p2ajqxe99nc0h8m/input.pdf
image_raw.png (output): https://www.dropbox.com/s/4cni4w7mqnmr0t7/image_raw.png
step 2: Stamp every PNG file (using a third party tool ..)
image_stamped.png (output): https://www.dropbox.com/s/3ryiu1m9ndmqik6/image_stamped.png
step 3: Reconvert PNG files into one PDF file
cmd: convert -resize 1240x1753 -units PixelsPerInch -density 150x150 image_stamped.png output.pdf
output.pdf (output): https://www.dropbox.com/s/o9y0jp9b4pm08ci/output.pdf
The output file of the third step shal be "theoretically" the same as the input file in step 1 (plus the stamp on it) but it's not, the file is somehow blurry and it turns to be unreadeable for humans after faxing it since blurred pixels wouldnt pass through fax wires even if you may see no difference between input.pdf and output.pdf, try zooming in and you'll find that text characters are blurred on its edges.
What is the best parameters to play with at input (step 1) or output (step 3) ?
Thanks !

Answer 1

You are using anti-aliasing (TextAlphaBits=4). This 'smooths' the edges of text by introducing grey pixels between the black pixels of the text edges. At low resolutions (such as displays) this prevents the 'jaggies' in text and gives a more readable result. At higher resolutions its value is highly debatable.
Fax is a 1-bit monochrome medium, so the grayscale values have to be recreated by dithering. As you have discovered, this is not a good idea in a limited resolution device as it leads to a loss of sharpness.
I believe that if you remove the -dTextAlphaBits=4 you will see an immediate improvement. I would also suggest that you remove the GraphicsAlphaBits as well, since this will have the same effect on linework.
If you believe that you still want anti-aliasing you could try reducing the aggressiveness, you currnetly have it set to 4, try reducing it to 2.
Regarding the other comments;
Kurt is quite correct, as is fourat, and I'm afraid MarcB is mistaken, the -r400 sets the resolution for rendering, in dots per inch. If only one number is given it is used for both x and y resolution. It is possible to produce a fixed size raster using Ghostscript, but you use the -dFIXEDMEDIA with -sPAPERSIZE switches or the -g switch which also sets FIXEDMEDIA automatically.
While I do agree with yms and Kurt that converting the PDF to a bitmap format (PNG) and then back to PDF will result in a loss of quality, if the final PDF is only used for transmission via fax, it doesn't matter. The PDF must be rendered to a fax-resolution bitmap at some point in the process, its not a big problem if its done before the stamp is applied.
I don't agree with BitBank here, converting a vector representation to bitmap means rasterising it at a particular resolution. Once this is done, the resulting image cannot be rescaled without loss of quality, whereas the original vector representation can be as it is simply rendered again at a different resolution. Image in PDF refers to a bitmap, you can't have a vector bitmap. The image posted by yms clearly shows the effect of rendering a vector representation into an image.
One last caveat. I'm not familiar with the other tools being used here, but two of the command lines at least imply 'resize'. If you 'resize' a bitmap then the chances are that the tool will introduce the same kinds of artefacts (anti-aliasing) that you are having a problem with. Onceyou have created the bitmap you should not alter it at all. Its important that you create the PNG at the correct size in the first place.
And finally.....
I just checked your original PDF file and I see that the content of the page is already an image. Not only that its a DCT (JPEG) image. JPEG is a really poor choice of format for a monochrome image. Its a lossy compression format and always introduces artefacts into the image. If you open your original PDF file in Acrobat (or similar viewer) and zoom in, you can see that there are faint 'halos' around the text, you will also see that the text is already blurry.
You then render the image, quite probably at a different resolution to the original image resolution, and at the same time introduce more blurring by setting -dGraphicsAlphaBits. You then make further changes to the image data which I can't comment on. In the end you render the image again, to a monochrome bitmap. The dithering required to represent the grey pixels leads to your text being unreadable.
Here are some ways to improve this:
1) Don't convert text into images like this, it instantly leads to a quality loss.
2) Don't compress monochrome images using JPEG
3) If you are going to work with images, don't keep converting them back and forth, work with the original until you are done, then make a PDF file from that, if you really must.
4) If you really insist on doing all this, don't compound the problem by using more anti-aliasing. Remove the -dGraphicsAlphaBits from the command line. You might as well remove -dTextAlphaBits as well since your files contain no text. Please read the documentation before using switches and understand what it is you are doing.
You should really think about your workflow here. Obviously we don't know what you are doing or why, so there may well be good reasons why some things are not possible, but you should try and avoid manipulating images like this. Because these are not vector, every time you make a change to the image data you are potentially losing information which cannot be recovered at a later stage. By making many such transformations (and your workflow as depicted seems to perform as many as 5 transformations from the 'original' image data) you will unavoidably lose quality.
If possible retain everything as vector data. When it is unavoidable to move to image data, create the image data as you need it to be finally used, do not transform it further.

Answer 2

I've had a closer look at the files you provided, see here:
So, already the first image (image_raw), the result of the mogrify resize command, is fairly blurry at 1062x1375. While the blurriness does not get worse in the second image (image_stamped) which is the result of the third-party tool, the third image (extracted from your output.pdf), i.e. the result of that convert command, is even more blurred which is due to the graphic being resized (which is something you explicitly tell it to do).
I don't know at which resolution your fax program works, but there is more quality loss still, at least due to 24 bit colors to black-and-white transformation.
If you insist on the work flow (i.e. pdf->png->stamped png->pdf->fax) you should
in the initial rasterization already use the per-inch resolution your rastered image will have in all following steps (including fax transmission),
refrain from anti-aliasing and use of alpha bits (cf. KenS' answer), and
restrict the rasterized image to the colorspace available to the fax transmission, i.e. most likely black-and-white.
PS As KenS pointed out, already the original PDF is merely a container for an image (with some blur to start with). Therefore, an alternative way to improve your workflow is to extract that image instead of rendering it, to stamp that original image and only resize it (again without anti-aliasing) when faxing.