Ghostscript renders ugly text

Question

I'm trying to add the capability to render LaTeX equations to a project I'm working on. To do so, I use XeLaTeX to create a PDF file, which I then render to a (transparent) 96dpi-PNG using Ghostscript.

I'd like to have the rendered LaTeX blend in with the rest of the text (which is rendered using standard .NET GDI+ methods, but that's off-topic), but I can't get a reliably "good" text rendering: the output always looks somehow blurry or otherwise "bad".

Example:

Comparison

From left to right, the same (small) PDF rendered at 96dpi with Ghostscript, Photoshop, and TexWorks (which I understand uses Ghostscript internally).

The command I use to run Ghostscript is the following:

"C:/Program Files (x86)/gs/gs9.09/bin/gswin32c.exe" \
      -q -dQUIET -dSAFER -dBATCH -dNOPAUSE -dNOPROMPT \
      -dMaxBitmap=500000000 -dAlignToPixels=1 -dGridFitTT=2 \
       "-sDEVICE=pngalpha" -dTextAlphaBits=4 \
      -dGraphicsAlphaBits=4 "-r96" -dFirstPage=1 -dLastPage=1 \
      -sOutputFile="output.png" "input.pdf"

(which I actually pretty much copied from the command ImageMagick calls when converting a PDF file, but that's another story). I tried changing any of the relevant options (dAlignToPixels=0, dGridFitTT=0/1/2, dTextAlphaBits=2/4 [or without this parameter altogether]) and I even tried to render the PDF to 4 times the resolution and then downscale it, without any noticeable improvement.

Yet, I'm sure there must be some way of decently rendering the PDF with Ghostscript (since TexWorks does), although I'm unable to find it.

Any hint? The PDF is this one.

Kurt Pfeifle Kurt Pfeifle · Accepted Answer · 2014-11-15T11:26:23

You could try to render your PDF at a higher resolution. 96dpi just isn't enough for text with 11 pt size.

If you use 192dpi and then scale the display of the resulting image to 50% (wherever you use the PNG), these parts should still appear in the same size as befor, but with a higher resolution. What used to be a 4x7 pixels 's' should now be a 8x14 pixels 's'...

Update

Ok, since my explanation seems to have been not comprehendible enough for the OP, here's the deal.

Generate a PDF file containing the word "Test", using Ghostscript. In my case, it is Ghostscript v9.10:

gs                       \
 -o test.pdf             \
 -sDEVICE=pdfwrite       \
 -g230x100               \
 -c "/Helvetica findfont \
     11 scalefont        \
     setfont             \
     1 1 moveto          \
     (Test) show         \
     showpage"

From this PDF, generate 6 different images depicting the word "Test", using 6 different resolutions. The gs is still Ghostscript v9.10 (to be checked with gs -version):

for i in 1 2 3 4 5 6; do      \
    gs                        \
     -o t$(( ${i} * 96 )).png \
     -r$(( ${i} * 96 ))       \
     -sDEVICE=pngalpha        \
     -dAlignToPixels=1        \
     -dGridFitTT=2            \
     -dTextAlphaBits=4        \
     -dGraphicsAlphaBits=4    \
      t.pdf ;                 \
done

This will create the following PNGs, as confirmed by ImageMagick's identify command:

identify -format "%f :   %Wx%H pixels  --  %b filesize\n" t[1-9]*.png
  t96.png :   31x13 pixels  --     475B filesize
 t192.png :   61x27 pixels  --     774B filesize
 t288.png :   92x40 pixels  --    1.1KB filesize
 t384.png :  123x53 pixels  --   1.43KB filesize
 t480.png :  153x67 pixels  --   1.76KB filesize
 t576.png :  184x80 pixels  --   2.01KB filesize

Create a sample LaTeX document and embed the different images side by side and/or line by line. Here is my sample code:

\begin{document}
Test
\includegraphics[height=7.5pt]{t96.png}
\includegraphics[height=7.5pt]{t96.png}
\includegraphics[height=7.5pt]{t192.png}
\includegraphics[height=7.5pt]{t288.png}
\includegraphics[height=7.5pt]{t384.png}
\includegraphics[height=7.5pt]{t480.png}
\includegraphics[height=7.5pt]{t576.png}
Test\\

{}

Test <== real text

\includegraphics[height=7.5pt]{t96.png}  <-- 96 dpi figure

\includegraphics[height=7.5pt]{t192.png} <-- 192 dpi figure

\includegraphics[height=7.5pt]{t288.png} <-- 288 dpi figure

\includegraphics[height=7.5pt]{t384.png} <-- 384 dpi figure

\includegraphics[height=7.5pt]{t480.png} <-- 480 dpi figure

\includegraphics[height=7.5pt]{t576.png} <-- 576 dpi figure

Test <== real text
\end{document}

Here is a screenshot (at 400% zoom) from the PDF created via LuaLaTeX from the above LaTeX code:

The line with the 8 "Test" words has actual text only in the first and the last word. The 6 words in between are images with 96, 96, 192, 288, 384, 480 and 576 dpi.

I hope you can see now clearly how scaling up your image generation to a higher resolution will result in better quality for your final PDF if you include the higher resolution images into your LaTeX code...

Ghostscript renders ugly text

2 Answers

Update