mirrored axes labels with imagesc

Hi,

when I execute the following code

a = rand(5,5)
a(3,3) = NaN
figure 
imagesc(a,'AlphaData',double(~isnan(a)))

I get mirrored axes labels and also the data cursor info is mirrored. Does anybody know why and how to avoid it?

I use MATLAB R2010b on Win XP.

Answer 1

[Jette]

The MathWorks support says this is a known problem of some ATI graphics devices. Updating the driver for the graphic device didn't help in my case. It is possible to switch the OpenGL renderer by

opengl('software')

This sovles my problem.

Thanks for your help!

Answer 2

[Jarrod Rivituso]

Are you referring to the y-axis being in reverse direction? If so, you can set it back...

a = rand(5,5)
a(3,3) = NaN
figure
imagesc(a,'AlphaData',double(~isnan(a)))
set(gca,'YDir','normal')

[Jette]

This helps. Thank you.

It seems that the y-axis itself (i.e. the data) as well as the labeling are mirrored. You can't read the text normally, but could it probably read it in a mirror if you hold it above the text.

Do you have an idea why this happens? I just tried it on another PC and there it works without the setting of the YDir

[Jarrod Rivituso]

Hmm... I'm not sure exactly what you mean. Do you mean that you want the first row of your matrix to actually be the last one? If so, check out the flipud function

If you're referring to the characters on the y-axis themselves as being backwards... I've never heard of anything like that. Sorry!

[Walter Roberson]

I *have* heard of the characters being backwards, but not for R2010b on XP. It was an earlier problem on a different OS -- OS-X is what comes to mind.

Oh, dang I'm good :) R2008a on Snow Leopard. http://www.mathworks.com/support/bugreports/585050

Answer 3

[Image Analyst]

Jette, it's doing exactly as designed. With images, the generally accepted convention is that the top line is 1 and the bottom row of pixels is row N. That is exactly what you are getting. See how your axis tick marks have "1" right at the center of your top row? That's the way it is supposed to be. Note that this is different that what you get by doing a line plot with plot() or a bar chart with bar().

[Jette]

I really meant that the characters of the labels are not readable as normal. The characters are sort of of top-down. You probably could read them with a mirror (I haven't tried it). So you can guess what the labels mean.

[Image Analyst]

Sorry but when I ran your code (with and without the y axis reversal) I don't observe that. The numbers always look 100% normal. Right side up and facing the right way - no mirror image at all. See if the Mathworks tech support can reproduce it. It may be some kind of weird pathological Java problem, in which case your only hope may be to contact Yair Altman (http://undocumentedmatlab.com/), or reinstall MATLAB and update Java.

[Yair Altman]

I've seen the label-reversal effect before. In my experience it sometimes happens after printing/saving figures. It is an internal Matlab bug. I am guessing it's due to some clash between Java and the graphics renderer (OpenGL etc.). I have no clue how to solve this Matlab bug other than to tell you not to use Matlab's print but rather use a screen-grabber.

I really think that if you can reproduce this effect you should email Matlab's customer service ([e-mail]) - Maybe they have a workaround, and if so then please post it here for all of us.

[Jan]

[e-mail], without "i".

Answer 4

[Jan]

Some OpenGL drivers have this known bug, that the text is mirrored. If you have tried already to install the newest graphics drivers, this is a workaround:

feature('UseGenericOpenGL', 1)

Because this is not completely documented, this might change with the Matlab release. In Linux it might be:

feature('UseMesaSoftwareOpenGL', 1)

See:

help opengl
opengl info

Answer 5

[Shreesha Vitthala]

The command "opengl AUTOSELECT" worked for me...

[Image Analyst]

Worked how? Do you mean it prevented the mirror image? Where exactly in the code did you insert this function?

By the way, since I answered this two and a half years ago, I did observe it on one person's computer. It was fine on my computer and another one but one scientist had the mirror image. I changed the renderer of the figure in GUIDE to zbuffer and that fixed the problem. It also changed the colors (e.g. yellow was more vivid) for some reason.