Why does Pcolor reduce the resolution of my n x m matrix

I am trying to make a plot of a matrix T (n x m) which corresponds to data at n-depth levels 'Z', evolving over m timestamps 't'. I always used to use imagesc(, but somehow this gives me a blank figure (???). t is slightly irregularly spaced, Z is regular.

Alternatively plotting imagesc(T) gives a nice plot, but I would prefer imagesc(t,Z,T) to work. Instead using pcolor() gives an ugly plot, which looks like an imagesc of n/6 columns only. What's going on?

[KSSV]

Try pcolor and then shading interp.

[Anneke ten Doeschate]

Thank you. that removes the lines but still very 'blocky'

Answer 1

[Star Strider]

The pcolor function uses the edges to define the patch colours, so it is ‘trimmed’ by one row and one column.

Use the image function instead.

[Image Analyst]

That's why I never use pcolor - it chops off the last row and last column. I use imshow() rather than image().

[Star Strider]

Good point.

I was thinking imshow and wrote image.

[Anneke ten Doeschate]

Thank you. Why do people then say pcolor is more 'powerful'? See the difference between plotting the same dataset (70x48 matrix) below ,1st with pcolor, 2nd with imagesc (got that one working after all)

[Star Strider]

Our pleasure.

I don’t know.

Image Analyst and I generally don’t use pcolor for the reasons we described, and use imshow (or imagesc) instead. I’m sure pcolor has its uses, just not in anything I routinely do.

[Walter Roberson]

"Why do people then say pcolor is more 'powerful'?"

image() and imagesc() and imshow() position the pixels equal distance apart, which is often what is desired. If you provide vectors of position information to any of these, it ignores all but the first and last x and y positions and uses the first and last as coordinates of the center of the bottom left and top right pixels. If you have more screen pixels than data pixels, then data pixels are just replicated to the screen (however, in R2014b and later, GraphicsSmoothing might do some anti-aliasing.)

pcolor() on the other hand accepts full 2D matrices of X and Y which are not required to be equally spaced at all, and interpolates color between those locations. Because the pixels need not be evenly spaced, there could be gaps where the data was sparser than the screen pixels, and other places where the data was denser than the screen pixels; it is all handled smoothly.

Basically, pcolor creates a colored surface plot and looks down on it from the top, so you can use pcolor to do texture mapping.

[Adam]

If you have irregularly space data I would suggest interpolating it onto a regular grid.

Personally I always use imagesc (and have only ever heard of pcolor from questions in the forum and never really knew what it did or why since its name does not convey to me anything that would suggest I'd want to plot an image with it!), but it depends what you are plotting and what you want to get from it.

I work with images that are recorded data rather than pictures (as you appear to be doing) so imagesc does exactly what I want, stretching the colourmap (though admittedly I usually over-ride this myself anyway). I have never had any desire to use imshow, but I know other people like it and I guess for photograph/picture type images it works fine and you don't have to map your data via a colourmap (though images either come with a colourmap or are greyscale or true RGB anyway so that is not difficult).

[Walter Roberson]

"If you have irregularly space data I would suggest interpolating it onto a regular grid."

pcolor / texture mapping is more efficient, in that it allows you to offload that work to your graphics hardware. That can make a significant difference if you are animating by updating coordinates.