Skip to content

Problems with graphical devices, e.g., png(), pdf(): blurry graphical output

5 messages · Y-H Chen, Brian Ripley, Martyn Plummer

Y-H Chen
# 
On my current home system, I am getting undesirable output from
graphical devices such as png() and pdf(). The graphical output is
blurry. I haven't experienced the problem on other systems. As you
will see from the attached text file (more information on this file
below), the problem does not occur when type='Xlib' is forced. The
blurriness is more severe with bitmap output (yes, I am viewing the
bitmap files at 100%), but occurs with pdf output as well.

Software details: Fedora 10, with at least the following packages:

-- R, R-core, R-devel
-- cairo, cairo-devel
-- pixman, pixman-devel
-- libpng, libpng-devel
-- poppler

Everything is "current" and updated via Fedora's repository. R was
installed via Fedora's repository.

I've attached some commands and output in a text file. This file includes:

(1) hardware information
(2) information about my R installation
(3) code for simple R graphics, with comments re output, plus URLs for
the corresponding graphical output

Any advice would be really appreciated.
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: R-graphics.txt
URL: <https://stat.ethz.ch/pipermail/r-help/attachments/20081216/4aeeb875/attachment.txt>
Brian Ripley
# 
Your PDF problems indicate a broken viewer.  How were you viewing PDF?

You have also not told us how you view PNG, but you would expect 
anti-aliased output to be blurry when viewed at 100% (or more).  You need 
to be careful not to have anti-aliasing turned on in the viewer as well as 
in the file producer.  Note that R does give you lots of options to tune 
the output to your intended use of it, so you have no cause to complain if 
you use an inappropriate set.

Do remember that you cannot say 'the graphical output is blurry': it is 
just a binary file.  Far too often, useRs blame R for issues in the 
viewers they use.
On Tue, 16 Dec 2008, Y-H Chen wrote:

            

      
      
      
    

        

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Y-H Chen
    

    

      
      #
    

  


  

      
On Wed, Dec 17, 2008 at 12:54 AM, Prof Brian Ripley

        
<ripley at stats.ox.ac.uk> wrote:

            

      
      
      
    

        
You are absolutely correct about the PDF files; I've since checked the
PDF files in other viewers and have not been able to reproduce the
problem. There was definitely something wrong with my default viewer.
I am most certainly embarrassed.

As for the PNG files: I've viewed the PNG files on two different
systems, in various viewers. On a Fedora system in Eye of Gnome,
Firefox, Epiphany, GIMP, and gThumb. And, on a Windows system in
Paint, Firefox, Internet Explorer, and GIMP. I see blurriness in all
of these viewers for all of the files I originally claimed difficultly
with (as mentioned in my last message, I don't see blurriness in the
images produced via type='Xlib'). I'll do more tests on other systems
once I get the chance, but that's what I see at the moment. The files
are linked to in the text file I provided, and you are all invited to
check those out if you are interested.

  
  


  


      

    

    

      
      

        


  

        

      Martyn Plummer
    

    

      
      #
    

  


  

      
The artefacts that you see are a normal result of using bitmap graphics
devices.  I have tried to explain these below:

I have looked at your figures in Eye of Gnome, with anti-aliasing turned
off (Menu Edit/Preferences; Tab "Image View"; option "Smooth images when
zoomed").  I recommend that you do the same.

            

      
      
      
    

        
Cairo works in real (double precision) coordinates. But the line must be
converted to bitmap to be displayed. When this is done without
anti-aliasing, it is quite possible for a thin horizontal or vertical
line to pass in-between the points on a grid that are sampled to form
the bitmap image, and hence disappear.

            

      
      
      
    

        
With anti-aliasing, a horizontal or vertical line may appear as a 1
pixel wide black line, but is more likely to appear as a 2 pixel wide
grey line.  When two such grey lines are over-plotted, they will create
a darker grey line. The overplotted line also appears thicker, but this
is an optical illusion.

            

      
      
      
    

        
Xlib works in integer coordinates. When a line is plotted in Xlib, the
start and end coordinates are cast to integer before plotting. Hence
horizontal/vertical lines will always appear, with a width of 1 pixel
and overplotting does not change the appearance of a line.

Martyn

        
On Wed, 2008-12-17 at 01:36 -0800, Y-H Chen wrote:

            

      
      
      
    

        

      
      
    

        
-----------------------------------------------------------------------
This message and its attachments are strictly confidenti...{{dropped:8}}

  
  


  


      

    

    

      
      

        


  

        

      Y-H Chen
    

    

      
      #
    

  


  

      
On Wed, Dec 17, 2008 at 3:06 AM, Martyn Plummer <plummer at iarc.fr> wrote:

            

      
      
      
    

        
Thanks very much for your explanations, MP; they were quite informative!!

I recognize that others may feel differently, but to me, the "default"
PNG being produced by my system -- e.g.,
http://www.piccdrop.com/images/1229495327.png -- is not ideal for
presentation on the web. E.g., the image mentioned in the last
sentence is blurry at 100% in every web browser and image viewer I've
tried (even when I turn off aliasing in Eye of Gnome, as MP suggests).

Given my feelings about the blurriness, my question is: how can I
produce a non-blurry image via png()? I recognize that I can do so via
png('plot.png',type='Xlib') but, I am wondering if there are solutions
that don't involve Xlib.

I guess my second question (if anybody has patience for it) is: what
is the philosophy behind the current behavior? I was able to find some
of that here:

http://www.cairographics.org/FAQ/#sharp_lines

... but I am wondering if somebody could elaborate on this philosophy
in relationship to R and statistical graphics, or point me to some
links that do so. The link above does suggest that sharp single-pixel
lines are possible via Cairo. So: are sharp single-pixel lines
possible via Cairo in R, and if so how (this of course the question
I've asked above)? And if not, why?

(Just to be clear: I'm not saying the current implementation is
necessarily "bad." At present, I'm just trying to understand more
about why it is the way it is.)