MATLAB Answers

Dario
0

imagesc: how to set NaN as white color

Asked by Dario
on 18 Dec 2017
Latest activity Commented on by Adam
on 7 Jun 2019
I need to set NaN values as white color using imagesc. I attach the matrix which is a 1200x1200 matrix with different values equal to NaN.

  0 Comments

Sign in to comment.

2 Answers

Answer by Adam
on 18 Dec 2017
Edited by Adam
on 7 Jun 2019

doc colormap
allows you to define whatever colourmap you want. You can tag a white point on the front of e.g. a Parula colourmap of any size you want, e.g.
cmap = [0 0 0; parula(6)];
In order to make sure only NaNs take the white value though you would likely have to do some messing around with caxis. It isn't easy to define a colourmap for many values where you want precisely one to have one specific colour on the colourmap, unless you map it to true RGB, in which case you can do whatever you want.

  5 Comments

Try it. My guess is that the bottom seventh of the pixel values (pixel values up to one seventh of the range, starting from the min up to the the min plus one seventh of the max-min range) will all show up as white.
I think you're right. I ended up using jet(200) so that a smaller portion of the cmap is white, I think this works. Thanks!
Adam
on 7 Jun 2019
Your data will be mapped linearly to your colourmap, so the bigger the colourmap the smaller the range of values mapped into each bin. If you have a 256-element colourmap then the bottom 1/256 ( 0.39% ) of values will be mapped to the first bin of the colourmap. Nans will also get mapped here, whatever size your colourmap and range your data.
This is why I said "It isn't easy to define a colourmap for many values where you want precisely one to have one specific colour on the colourmap". If you are dealing with a uint8 image it is quite easy as you can have 1 bin per pixel value and just add an extra bin for the NaNs.
As also mentioned, you can play around with the caxis limits to achieve the right effect, but it all depends on your data.
e.g.
a = rand( 10 );
a(3,4) = NaN;
figure; imagesc( a );
colourmap( [0 0 0; parula(256)] )
If you do only that then quite likely you will have multiple black pixels instead of just 1 (this is random data though so it depends on the particular randon numbers you get. In my run I ended up with 4 black squares)
Now you can try:
caxis( [-0.001 1] )
and some or all of the non-NaN values will now take on a colour from the true colourmap instead of white. But if you have a very small value you may need to go for more like
caxis( [-0.01 1] )
or
caxis( [-0.1 1] )
by trial and error (or it can be mathematically calculated, I just couldn't be bothered)

Sign in to comment.


Answer by Image Analyst
on 7 Jun 2019

Have you tried this solution:
yourImage(isnan(yourImage)) = 255;

  0 Comments

Sign in to comment.