How to interpolate and smooth across values in a matrix, while ignoring NaN values?

Hi there!

I am having some trouble figuring out how to interpolate across a matrix while ignoring NaN values. I have matrix data where each cell represents information from a single microphone, some of the microphones in the array malfunctioned, returning 0 values. I don't want these microphones contributing 0s to my interpolation, as the 0s don't represent real data, but rather nonfunctioning microphones. I set these malfunctioning microphones to NaN. My matrix is called 'scan' and looks something like this.

.3 .4  .3 .2 NaN .2
.4 .4  .3 .4 .2  .2
.3 NaN .5 .2 .4. .4
.3  .3 .4 .2 .2  .1

I need to interpolate across values in this matrix but I want the interpolation to ignore the NaN values, interpolating across them but without receiving them as input. Below is the interpolation code before I tried to have it ignore the NaN values

  scan2 = scan'; %tranposes scan to scan2
  interpImage = scatteredInterpolant(micsX(:),micsY(:), scan2(:), 'natural'); %interpolates scan2 onto fine grid, calls it enInterp
  enInterp = interpImage(meshImageX, meshImageY);
  K = 0.125*ones(1, 1); %smooths the interpolated data
  enInterpSmooth = conv2(enInterp, K, 'valid');
  enInterpSmooth = enInterpSmooth'; %puts enInterpSmooth back to the original orientation

The above code returns the interpolation with huge chunks of NaNs surrounding where the NaN microphones were.

I added the below code to try and get an interpolation that ignored the NaN values, but it still returned large amounts of NaNs.

% Replace NaN values in enInterpSmooth with interpolated values
nanIndices = isnan(enInterpSmooth);
enInterpSmooth(nanIndices) = enInterp(nanIndices); % Replace NaNs with interpolated values

Any recommendations for how to remedy this problem would be much appreciated!!

Thank y'all so much!

Answer 1

[Voss]

Avoid putting NaNs into the scatteredInterpolant:

idx = ~isnan(scan2);
interpImage = scatteredInterpolant(micsX(idx),micsY(idx), scan2(idx), 'natural'); %interpolates scan2 onto fine grid, calls it enInterp
enInterp = interpImage(meshImageX, meshImageY);

[Jade]

Thank you so much, you helped me out yet again! :)

[Voss]

You're welcome!

Answer 2

[Mathieu NOE]

hello

just for your information, if you have a recent matlab release you could use fillmissing2 function

scan = [0.3 0.4  0.3 0.2 NaN 0.2
0.4 0.4  0.3 0.4 0.2  0.2
0.3 NaN 0.5 0.2 0.4 0.4
0.3  0.3 0.4 0.2 0.2  0.1]

scan = 4x6

0.3000 0.4000 0.3000 0.2000 NaN 0.2000 0.4000 0.4000 0.3000 0.4000 0.2000 0.2000 0.3000 NaN 0.5000 0.2000 0.4000 0.4000 0.3000 0.3000 0.4000 0.2000 0.2000 0.1000

<mw-icon class=""></mw-icon>

<mw-icon class=""></mw-icon>

scan2 = fillmissing2(scan,"v4")

scan2 = 4x6

0.3000 0.4000 0.3000 0.2000 0.1961 0.2000 0.4000 0.4000 0.3000 0.4000 0.2000 0.2000 0.3000 0.4308 0.5000 0.2000 0.4000 0.4000 0.3000 0.3000 0.4000 0.2000 0.2000 0.1000

<mw-icon class=""></mw-icon>

<mw-icon class=""></mw-icon>

[Mathieu NOE]

you can also use inpaintn available on the fex

Inpaint over missing data in 1-D, 2-D, 3-D,... ND arrays - File Exchange - MATLAB Central (mathworks.com)

the result may slightly differ as we are not using the same approach

scan2 =

0.3000 0.4000 0.3000 0.2000 0.1599 0.2000

0.4000 0.4000 0.3000 0.4000 0.2000 0.2000

0.3000 0.3943 0.5000 0.2000 0.4000 0.4000

0.3000 0.3000 0.4000 0.2000 0.2000 0.1000

[Mathieu NOE]

I have one more suggestion (John would have been unhappy if I had forgotten about his work !!)

inpaint_nans - File Exchange - MATLAB Central (mathworks.com)

scan2 =

0.3000 0.4000 0.3000 0.2000 0.1167 0.2000

0.4000 0.4000 0.3000 0.4000 0.2000 0.2000

0.3000 0.4222 0.5000 0.2000 0.4000 0.4000

0.3000 0.3000 0.4000 0.2000 0.2000 0.1000