Skip to content

png() and image()

7 messages · Henrik Bengtsson, Wolfgang Huber, Hin-Tak Leung +1 more

#
Hi,

I try to create PNG images of a certain size where each pixel
intensity corresponds to exactly one probe signal in an Affymetrix
array.  I try to use png() and image() with zero margins to do this.
Example:

z <- matrix(1:15, nrow=45, ncol=30)
png("large.png", height=nrow(z), width=ncol(z), bg="red")
par(mar=c(0,0,0,0))
image(z, col=gray.colors(16), axes=FALSE)
dev.off()

z <- matrix(1:15, nrow=5, ncol=3)
png("tiny.png", height=nrow(z), width=ncol(z), bg="red")
par(mar=c(0,0,0,0))
image(z, col=gray.colors(16), axes=FALSE)
dev.off()

The problem is that on WinXP the very bottom row and the very right
column of pixels in red.  Trying on Linux, it is only the very right
column that is red.  See attached images (you might have to zoom in to
see it).  I try to do this in R v2.3.1.  The same effect is seen if
the jpeg() device is used.

When rescaling, the same effect is seen (the red border effect is one
pixel wide), e.g.

z <- matrix(1:15, nrow=45, ncol=30)
png("large5.png", height=5*nrow(z), width=5*ncol(z), bg="red")
par(mar=c(0,0,0,0))
image(z, col=gray.colors(16), axes=FALSE)
dev.off()

I might be asking for something that is not supported, but is there a
way around this?  It is a problem, because I wish to tile the images
in an HTML page.

Thanks

Henrik
-------------- next part --------------
A non-text attachment was scrubbed...
Name: WinXP-large.png
Type: image/png
Size: 141 bytes
Desc: not available
Url : https://stat.ethz.ch/pipermail/r-devel/attachments/20060720/ca12637e/attachment.png 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: WinXP-tiny.png
Type: image/png
Size: 113 bytes
Desc: not available
Url : https://stat.ethz.ch/pipermail/r-devel/attachments/20060720/ca12637e/attachment-0001.png 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: Linux-large.png
Type: image/png
Size: 148 bytes
Desc: not available
Url : https://stat.ethz.ch/pipermail/r-devel/attachments/20060720/ca12637e/attachment-0002.png 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: Linux-tiny.png
Type: image/png
Size: 118 bytes
Desc: not available
Url : https://stat.ethz.ch/pipermail/r-devel/attachments/20060720/ca12637e/attachment-0003.png 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: WinXP-large5.png
Type: image/png
Size: 317 bytes
Desc: not available
Url : https://stat.ethz.ch/pipermail/r-devel/attachments/20060720/ca12637e/attachment-0004.png 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: WinXP-tiny5.png
Type: image/png
Size: 173 bytes
Desc: not available
Url : https://stat.ethz.ch/pipermail/r-devel/attachments/20060720/ca12637e/attachment-0005.png
#
Hi Henrik,
this does not directly answer your question, but you might be able to 
use the function write.image in the package EBImage (Bioconductor) to 
write matrices into image files (e.g. PNG or JPEG), this might be more 
flexible and also faster than what you're trying to do.
  Best wishes
  Wolfgang
------------------------------------------------------------------
Wolfgang Huber  EBI/EMBL  Cambridge UK  http://www.ebi.ac.uk/huber
Henrik Bengtsson wrote:
#
Hi Wolfgang,

thanks for this - I took a look the EBImage webpage
[http://www.ebi.ac.uk/~osklyar/projects/EBImage/] and it looks very
nice.  Are there any plans for Windows support too?  I am looking for
a cross-platform solution.

Cheers

Henrik
On 7/21/06, Wolfgang Huber <huber at ebi.ac.uk> wrote:
#
By changing
   png("large.png", height=nrow(z), width=ncol(z), bg="red")
to
   png("large.png", height=nrow(z), width=ncol(z), bg="blue")
I can get the right column (on linux) to change to blue.

So it seems to be an alignment off-by-one error with the background
colour showing through, so you might be able to work around it
by changing the alignment of the margins.

Incidentally I am working on the AffyMetrix 500k chip as well
(http://www.wtccc.org.uk/) but I have already written a non-R C/C++
tool to dump all the probe signal of a whole array as an image
(not intended for html, obviously, for 6.5 million probes); and
another one for drawing the schematics of the locations of
probes associated with a specific snp as a postscript vector diagram.
I will probably be working on combining the two - i.e. overlaying
an image with a vector diagram next week. (which means I hope to get
an image of intensities with some cicles on top, highlighting the
effect on the intensities to a specific snp due to blemishes).

Hin-Tak Leung
Henrik Bengtsson wrote:
2 days later
#
Henrik Bengtsson wrote:
Hi Henrik,

there are very strong intentions to also make this available as a
Windows precompiled package and/or provide installation instruction for
Windows. Until now however, in spite of some effort, we have failed.

As I understand it, a problem seems to be different dll formats of  MS
Visual C++ (in which ImageMagick is provided) and MinGW.

If anyone is interested in helping with this, they could have a look at
the source package on Bioconductor, and contact us for details.

 Best wishes
 Wolfgang

  
    
#
Wolfgang Huber wrote:
It is probably easier to port ImageMagick to MinGW than to try to
link mingw-compiled EBImage against MSVC++ compiled imagemagick.
(Imagemagick does run under cygwin, I believe, so it is not
entirely hostile to gcc on windows).

HTL
#
There are no such things: the Windows DLL format is one DLL format, and R
uses VC6-compiled DLLs for Tcl/Tk and iconv, for example.  And other
packages (e.g. RNetCDF, rgdal) take the same route.

The issues are usually to use the correct header files that mark entry
points as stdcall rather than cdecl as needed, and to link directly to
the DLL rather than via a VCC-specific import library (which may often
work, but equally often gets the linkage wrong.
That would run contrary to most experience with Windows porting.