Antialiasing and subpixel plotting

1
Greetings,

I am switching over from the OS 9.2 Psychtoolbox to the OsX version
for the Macintosh. How can I antialias diagonal lines and get
subpixel resolution on my stereo images (I'm using a split screen and
mirrors at the moment)? I used to do it by drawing big pictures in
the OffScreenWindow and shrinking them down to the OnScreenWindow
using the CopyWindow command. The DrawTexture function looks like
the thing to use but I don't know what a "texturePointer" is.

Many thanks in advance.

Dr. Philip Grove
School of Psychology
The University of Queensland
St. Lucia, QLD 4072
Australia
p.grove@...
2
Hi Jenny,

version 1 of your code with Screen('MakeTexture') should work if you
provide the dstRect parameter explicitely to either
Screen('DrawTexture') or Screen('CopyWindow'). If you don't provide
it, PTB will center your texture in the onscreen window, so if your
image is bigger than the screen, you'll just see the center-portion of it.

If you set dstRect = rect, as returned by Screen('OpenWindow'), PTB
will rescale (downsample by use of bilinear filtering) the image to
fit the screen and you'll get what you want.

Method 2: Drawing into offscreen window has currently one limitation.
It does not allow for offscreen windows greater than the screen, so
this may only work for small stimuli.

But here is some new way of getting subpixel resolution and
Anti-Aliasing without any special coding:

I've updated the PTB beta a few minutes ago: Now you can pass a new
optional parameter 'multiSample' to Screen('OpenWindow') -- see its
updated online help. If you set multiSample to a value greater than
zero, then PTB will ask your gfx hardware to perform automatic
anti-aliasing on all drawn content, where the value of 'multiSample'
specifies the number of samples to use for computing each pixels final
color value. Values supported by todays hardware are, depending on the
age of your hardware, screen resolution and amount of video memory on
the card: 2, 4, 8 and (on very new cards at low resolution) 16. If you
ask for a value greater than the one supported by your hardware, PTB
will simply use the highest supported value. The Anti-Aliasing mode is
reported to the Matlab window. This method of brute force
Anti-Aliasing obviously comes at a cost: The much higher amount of
required video memory and memory bandwidth will significantly reduce
the drawing speed, so only use it if you need it.

A good beginners intro on how this works can be found at:

<http://en.wikipedia.org/wiki/Supersampling>

best,
-mario
3
Hi Mario,

Thanks so much for that quick and incredibly helpful reply.

I've downloaded the new PTB beta and had a go with the new multisample.
It works beautifully and does just what I want. (At least, with
multisample=4 it does -- with multisample=8, I got a red flashing
screen, a yellow exclamation mark and lots of warnings about
synchronisation failure. But that's fine; I know not to use that in
future). Thanks so much for adding this.

I did also try your other solution, setting dstRect to rect, but this
didn't work for me. The code I used was

> % Set up screen
> [window,rect]=Screen('OpenWindow',0);
> nx = rect(3); ny = rect(4);
> white=WhiteIndex(window); black=BlackIndex(window);
> x = [1:nx]; y = [1:ny]; [x2,y2] = meshgrid(x,y);
>
> % Generate a simple image that's twice as big as my screen along
> both axes. It's black with part of a white circle.
> expandfactor = 2;
> im=black*ones(ny*expandfactor,nx*expandfactor);
> im(x2.^2+y2.^2<500000)=white;
>
> % Now squish this image into my screen.
> textureIndex=Screen('MakeTexture', window, im);
> Screen('DrawTexture', window, textureIndex,rect*expandfactor,rect,1);
> % I also tried Screen('DrawTexture', window, textureIndex,[],rect,1);
> Screen('Flip', window);


Is this what you meant? This did give me an antialiased image, but it
was doubled vertically, ie the picture appeared twice down my screen,
and there was some other weird stuff too round the edges. Of course I
don't need this to work now you've provided the multisample function,
but I thought you might like to have the feedback.

Best wishes and very many thanks for your help,

Jenny

(using revision 363, beta version 1.0.6 on Windows XP).

--
Jenny Read Office: +44 191 222 7559
Royal Society University Research Fellow Mobile: +44 794 401 5796
Henry Wellcome Building for Neuroecology, Fax: +44 191 222 5622
University of Newcastle, Framlington Place mailto:J.C.A.Read@...
Newcastle upon Tyne, NE2 4HH, UK. http://www.staff.ncl.ac.uk/j.c.a.read



Mario Kleiner wrote:

> Hi Jenny,
>
> version 1 of your code with Screen('MakeTexture') should work if you
> provide the dstRect parameter explicitely to either
> Screen('DrawTexture') or Screen('CopyWindow'). If you don't provide
> it, PTB will center your texture in the onscreen window, so if your
> image is bigger than the screen, you'll just see the center-portion of it.
>
> If you set dstRect = rect, as returned by Screen('OpenWindow'), PTB
> will rescale (downsample by use of bilinear filtering) the image to
> fit the screen and you'll get what you want.
>
> Method 2: Drawing into offscreen window has currently one limitation.
> It does not allow for offscreen windows greater than the screen, so
> this may only work for small stimuli.
>
> But here is some new way of getting subpixel resolution and
> Anti-Aliasing without any special coding:
>
> I've updated the PTB beta a few minutes ago: Now you can pass a new
> optional parameter 'multiSample' to Screen('OpenWindow') -- see its
> updated online help. If you set multiSample to a value greater than
> zero, then PTB will ask your gfx hardware to perform automatic
> anti-aliasing on all drawn content, where the value of 'multiSample'
> specifies the number of samples to use for computing each pixels final
> color value. Values supported by todays hardware are, depending on the
> age of your hardware, screen resolution and amount of video memory on
> the card: 2, 4, 8 and (on very new cards at low resolution) 16. If you
> ask for a value greater than the one supported by your hardware, PTB
> will simply use the highest supported value. The Anti-Aliasing mode is
> reported to the Matlab window. This method of brute force
> Anti-Aliasing obviously comes at a cost: The much higher amount of
> required video memory and memory bandwidth will significantly reduce
> the drawing speed, so only use it if you need it.
>
> A good beginners intro on how this works can be found at:
>
> <http://en.wikipedia.org/wiki/Supersampling
> <http://en.wikipedia.org/wiki/Supersampling>>
>
> best,
> -mario
>
>
4
Hi Jenny,

beta updated again: Better error-checking, special case for
multisampling implemented, always selects high-quality filter,
stereomode 4 (the one for nView desktop spanning) improved to be
actually useful on a setup like yours.

Could you retest with this beta?

best,
-mario
5
It is pretty easy. Make a file called startup.m somewhere on the path
(on linux or mac the best place is ~/matlab, I'm afraid I'm not sure
where the best place is on windows). Startup is run everytime matlab
starts. Add this line to your startup.m:

addpath(genpath('~/matlab/Psychtoolbox'));

Again, this is path stuff for linux or mac, but shouldn't be hard to
translate it appropriately for windows. This command will put
everything in your Psychtoolbox folder on your path. Give your other
Psychtoolbox a different name, like Psychtoolboxbeta, and also put
this in your startup.m:

addpath(genpath('~/matlab/Psychtoolboxbeta'));

You can comment out the one that you don't want to use. Anytime you
want to switch, or after installing/updating the psychtoolbox, run
restoredefaultpath. Or you could just have this as the first command
in the startup.m This will get rid of anything extra on your path.
Then run startup. You could even do something fancy, like put an
optional argument to startup that lets you choose which Psychtoolbox
you want (make sure you have just one of the toolboxes as the default
though, so when matlab starts it just has one psychtoolbox on the
path).

Doing it this way means that you need to put everything you want on
your path in the startup, rather than using the gui, but this isn't
hard, since the addpath(genpath()) syntax adds subdirectories.

hope this helps,
cheers,
maria

On 7/22/06, Charles E. (Ted) Wright <cew@...> wrote:
> Hi -
>
> Having been monitoring this thread about the Windows support of the beta
> branch of the PTB, I am interested in trying it out. My question is how
> difficult it will be to have both on the same machine. To do this will I
> need to reset all of the entries for the PTB in the Matlab path each time I
> want to switch back and forth. I imagine that I could put together a
> function to do this automatically -- is that the idea?
>
> Thanks,
>
> Ted Wright
>
> On Fri, 21 Jul 2006, Mario Kleiner wrote:
>
> > You need to DownloadPsychtoolbox('beta',...). The 'stable' branch
> > doesn't support M$-Windows or any other feature added in the last 6
> > months yet. If you did that, then a 'which Screen' should show you
> > that Screen.dll is installed inside the downloaded PTB, otherwise you
> > need to reset your Matlab path properly.
> >
> > -mario
>
>
>
> Post your message to: psychtoolbox@yahoogroups.com
> Please indicate OS9, OSX, or WIN version, and include your full name.
> Denis Pelli, David Brainard, and Allen Ingling.
> http://psychtoolbox.org
>
> Yahoo! Groups Links
>
>
>
>
>
>
>


--
Maria Mckinley
Scientific Programmer
Shadlen Lab
Physiology and Biophysics
University of Washington
(206) 616-7107
parody@...