Contourf is plotting weird lines. How to remove them?

Hi,

I'm using hist3 to get bin centers and the elements in each bin. After this I use contourf to plot the data. Unfortunatelly the outputed contour figure has small vertical and horizointal black lines that don't seem to make sense. Please see lines inside red circles for examples:

My code is as follows:

edges = {0:0.2:13 0:0.2:13};        % edges for hist3
insetPos = [.52 .53 .35 .35];  

[n,c] = hist3(X,'edges',edges);
figure
heatmap(n)                  % FEX
% contourmap
axes('Position',insetPos)
box on
contourf(c{1},c{2},n)

I belive the black lines are part of the contour lines, yet they appear where there are no boundries. If I remove the contour lines they also go away.

contourf(c{1},c{2},n,'LineColor','none')  % no contour lines

Any tip on how I could keep the contour lines without drawing the "intruding" lines?

[Walter Roberson]

can you attach X for testing?

[MDias]

Here is a .mat file with X. Thanks!

Answer 1

[Kelly Kearney]

When you let contour choose contour levels automatically, it doesn't always choose the levels you really want. In this case, it chooses integer values, which may not be the best way to contour all-integer data. You might have better luck getting the look you want if you specify between-integer levels, e.g.

contourf(c{1}, c{2}, n, -0.5:10.5)

[MDias]

Thanks Kelly!

Like Walter pointed out I didn't carefully at n...or at contourf for that matter. This does what I need.

Answer 2

[KSSV]

Use pcolor with shading interp instead of contourf.

[MDias]

Thanks for the suggestion! Unfortunatelly it also gets rid of the "true" contour lines, which I want to keep and the end result is less clean....

[KSSV]

On this use holdon and plot contour lines using countour.

[MDias]

The unwanted bits of the contour (small black lines highlighted above) are still there.

Answer 3

[Walter Roberson]

Look at n(24,9) and see that it is 1 there, corresponding to x=1.7, y=4.7; n(23,9) is also 1. That is an island of two 1's surrounded by 0's. With the contour levels happening to correspond to the integers, there should be an island there.

spy(n) and you will see there are a number of small islands.

[Walter Roberson]

if you take

n.*bwareafilt(n>0,[SIZE, inf])

then it would filter out islands smaller than SIZE pixels

[MDias]

Thanks Walter!

Now this makes sense, I did not look carefully at n...