Bottom and left edges of pcolor plot?

When I use pcolor to plot a colored surface based on values in a matrix, both the lower and right edge of the box around the axes do not show. I have tried different shading calls, turning the box on, and setting the layer to top. This happens in both R2012a with Mac OS and R2012b with Windows 7. What options are there for making those missing edges show up? Code that reproduces the problem is as follows:

C = rand(20) ;
pcolor(C)
shading interp
box on
set(gca,'Layer','top')

Answer 1

[Image Analyst]

That's what I've been telling unsuspecting people for years. It chops off a row and column. Unless you know that you need, really need, pcolor(), don't use it. Use image() or imshow() instead. I don't even like imagesc() because it applies some arbitrary colormap by default that is almost never helpful.

[Orion]

This is a very bad advice. image() and imagesc() DO NOT have the accuracy of pcolor. Similar to pcolor, image(X,Y,Z) accept X and Y position vectors, but it only cares about the minimum and maximum values, and then stretches the pixels in between to fill the axes. This absolutely fails when you have a nonmonotonic X or Y arrays.

In fact, the documentation for image() shows that the x-vector should be a two element array. But if you use a longer array, it still returns a plot without any error or warning. This is a major bug in the algorithm given that, based on your comment, people could mistakenly use pcolor and image interchangeably.

[Chad Greene]

Orion--funny you should mention that. Just last week I submitted a feature request for an input checker to make sure x,y coordinates with image and imagesc are either two-element endmembers, or have lengths corresponding to the number of columns & rows respectively, and are evenly spaced and monotonic. I've accidentally given bad inputs to imagesc many times without ever getting an error. MathWorks says they understand the issues and they're considering adding input checks in a future release.

[Image Analyst]

I didn't even understand his comment at all. I was just stating that if you use pcolor() to display an image, such as the poster's 20x20 "C" array, it will not show the lower and right edge of the array, as the poster stated, and you could use image() or imshow() to get that last row and column to show up.

C = rand(20);
subplot(1, 2, 1);
props = pcolor(C)
shading flat
box on
axis('on', 'image');
set(gca,'Layer','top')
title('pcolor() shows 19 columns', 'FontSize', 20);

subplot(1, 2, 2);
imshow(C);
axis('on', 'image');
title('imshow() shows 20 columns', 'FontSize', 20);

This is because the value for pcolor is shown at the edges of the pixels, not over the whole pixel like imshow(). But if you're looking at the whole image like the colored blocks were pixels, you're going to be missing one. imshow() displays it like there is a pixel block centered at the index as shown in the figure above, so you see all the pixels.

Neither me nor the poster was talking anything about X, Y, or Z, scaling, stretching, or color accuracy. Even if I were passing in X or Y to imshow() to get the tick marks to show up in calibrated units, I don't see why anyone would pass in an X or Y with non-monotonic values. That seems crazy. Why would you want X to ever go in reverse for part of the way with your image?

[Adrian]

Agree with Orion. Non-monotonic also means increasing in uneven steps and that is more common than you may think. If that is the case do NOT use image() or imagesc(), you would just be fooling youself as these functions only care about the first and last values to be used as limits, distributing the in-between pixels evenly, regardless if that's the case or not.

Answer 2

[Walter Roberson]

http://www.mathworks.com/help/matlab/ref/pcolor.html

When you use shading interp, each cell's color results from a bilinear interpolation of the colors at its four vertices.

Now consider that if you had a 2x2 array of data, there would only be one place that had four vertices: the point at which the four pixels meet. One location, one drawn cell. N x M locations, (N-1) x (M-1) drawn cells.

Answer 3

[Chad Greene]

Here's an example of the funny things pcolor, surf, and imagesc can do to your data.