Using a heatmap to display a large value range with detail

I have a matrix composed of a very large data set around ~1000000 elements.

The problem I am facing is that the first half of elements have a tiny value, with the data displaying differences of 1.0E-9, but this data is quite important to the analysis. The second half have difference ranges of 1.0E-3 to 0.5.

I have attempted to plot a heatmap, image and surf graph of the data but am unable to get a resolution good enough to display specific.

Is there a way to display the data that I need? Such that even the smallest change in value is represented just as well as the larger changes?

I have attached the images generated.

Answer 1

[Youssef Khmou]

HI, suppose your final data is R , try to plot it in dB:

 imagesc(10*log10(R));
 surf(20*log10(R));

If there are elements that approach zero, you can try to Rescale by multiplying by 1E+3 per example and try again with to visualize in dB (20log10(rescaledR)).

I hope that helps,

Your heatmap is result of some calculations or importation ?

[Fayzan Ahmed]

The heatmap is a result of calculations from software in Java, which is output to .data in a matrix format.

I then import this data to MatLab for analysis.

A new problem with taking the logarithm is that it introduces complex numbers and I cannot plot it. Is there a way around this?

[Fayzan Ahmed]

The reason is because negatives were in the matrix. This solution has helped massively. I have attached the new heatmap after taking a logarithm:

[Youssef Khmou]

hi, ok good

It looks like a problem that i saw somewhere , is that heat diffusion using Navier Stokes equations? At the end you are concerned about Real values as result, ok let is try to work on that : As R is your matrix,

1-try to visualize the imaginary part :

 imagesc(imag(10*log10(R)))

Is there any kind of symmetry between this and the real part?

2-try to visualize the modulus ( real®+imag®)^2 ) :

 imagesc(abs(10*log10(R)))

and try to rescale the result by / or * by some factor .

If one theses suggestions work , try to surf the result, insert a legend and see if really the result is correct, per example the Red region must have an altitude of 400 Kelvin .

to be continued

[Youssef Khmou]

Or you can try to Re-scale as "image Analyst" mentioned like the following :

 factor=max(-R(:));
 New_R=R+factor ;
 %Visualization with log:  now log will not produce C| values  .

Answer 2

[Image Analyst]

One common way to handle this is to take the log of the data before you send it into surf(), image(), or imshow(). That will compress the higher values and expand the lower values, allowing you to see the structure of your data over a greater range.

[Image Analyst]

Regarding your comment about taking the log and getting complex numbers. You can normalize your data to scale it between 1 and 100 so that you don't get values of 0 or less. The visualization should look virtually identical. You might want to look at mat2gray().