Replace repetitive values in a matrix with another value?

Hello there,

I'm new to Matlab so hopefully I can attempt to explain this! I have an image that has some interference that I am looking to get rid of. I have displayed each pixel in the image as a value from 0 to 255 in a matrix. The interference is in the form of repetitive values. So for example (and simplicity sake) one row of the matrix may look like:

1 2 3 3 3 3 3 4 5 3 3 3 6 3 7

Where the interference would be the string of 3 3 3 3 3 and 3 3 3 and the single 3 there would be valid.

What I would like to do is replace the repetitive values in the entire matrix with a nearest neighbor value say median of the pixel above and below and if that's not possible maybe just NAN. Would anyone know how to do this? Thanks!

[Ryan]

is the repetitive value known? is the repetitive value always the same for a given image (e.g. will the repeated value in the image always be a 3)?

Answer 1

[Sean de Wolski]

x = [1 2 3 3 3 3 3 4 5 3 3 3 6 3 7] data
idxrep = ~diff([-inf,x]) %pad with -inf and identify areas of zero change (~)

x(idxrep)=nan; %replace with NaN, or whatever

If you chose the NaN route and want an awesome way to fill them:

y = inpaint_nans(x)

[Andrei Bobrov]

x([false,diff(x)==0]) = nan;

[JL]

Hi, would it be possible to for this code to work on have more rows?

Answer 2

[Walter Roberson]

Convert the vector to double-precision data type (needed because integers cannot represent NaN.) diff() it. Replace the places the diff() was 0 with NaN.

Step after that: consider using the File Exchange inpaint_nans

Answer 3

[Jessica]

The problem that I can see with diff() is that it reduces the number of elements in each row by 1 because it takes the difference between adjacent values. I do not want the matrix size to change at all because it is an image I wish to analyze afterwards and pixel position is important.

[Sean de Wolski]

Pad with -inf

diff([-inf,x])

Then what Walter said.

[Jessica]

Sorry! I'm new to Matlab so what exactly would that do? Thanks for the responses!

Answer 4

[Jessica]

So I have

imgname = 'my tif file path'
imgmat = imread(imgname);

idxrep = ~diff([-inf,imgmat])
imgmat(idxrep) = NaN

and I get

Error using horzcat CAT arguments dimensions are not consistent.

Error in test (line 6) idxrep = ~diff([-inf,xd])

[Sean de Wolski]

Well, yes. My example was for a row vector. For a matrix you'll have to do a little more work. Specifically you will have to add a column vector to the front of the matrix (rather than a scalar) (this is the cause of the HORZCAT error). Second, you will have to specify which dimension you want to differentiate across.

x = repmat([1 2 3 3 3 3 3 4 5 3 3 3 6 3 7],10,1);
idxrep = [false(size(x,1),1) ~diff(x,1,2)]
x(idxrep) = NaN

[Jessica]

Ahh! Gotcha! Thanks! And thanks for the inpaint_nan link everyone! That was extremely helpful!