data correction

i have a matrix of topographic data. 1356*3039 there are some NaN cells in this matrix that I should remove them. for example like this: 928 926 932 939 970 938 937 936 940 965 949 931 940 928 957 1000 984 NaN 988 987 1108 1094 1121 1117 1068 1253 1285 1313 1226 1169

I want to create a rule to detect such cells and fill them with an interpolation of neighbor cells.

I can not use isnan function for all my data because this is a earth sea topographic data and earth data are also NaN.

Answer 1

[Clemens]

well in some way you will have to use isnan. You could use bwconncomp to find "lonely" Nans. Like in this example:

%%create some sample data
data = rand(10); % make a random sample
data(data>0.9) = NaN; % create some nans
data(1:6,1:6) = NaN; % make an island
c = bwconncomp(isnan(data))
s = cellfun(@numel,c.PixelIdxList)
single_nan_ind = s==1
single_nans = cellfun(@(x) ind2sub(x,size(data)),c.PixelIdxList(single_nan_ind),'UniformOutput',false)

The ones you are looking for would be in single_nans.

[Mahmood dandelion]

thank you

it works.

is there any way to find islands with 2 or more nan cells?

[Clemens]

Sure just fix the line:

single_nan_ind = s==1

this finds all groups with just one.

So single_nan_ind = s>5 would find islands with more than 5 nans.

I think you get the idea.

Answer 2

[Wolfgang Schwanghart]

Hi,

you can identify the NaNs using the function isnan.

I = isnan(dem); % where dem is the digital elevation model

The inpaint function by Damian Garcia can be used to fill the gaps.

http://www.mathworks.com/matlabcentral/fileexchange/27994-inpaint-over-missing-data-in-n-d-arrays

Cheers, W.

[Mahmood dandelion]

thank you

but I said that I cannot use isnan, because I have some data that should be 'nan'. I want to find the nan cells that all 8 cells around them were not nan and also data cells that have the same position due to nan cells.

I am trying to write a script that allow me to find cells that rounded by a range of data in special radius.

for example if a(i,j)=nan then a(i-1:i+1,j-1) & a(i-1:i+1,j+1) & a(i-1,j) & a(i+1,j) ((8 cells around i,j)) were all nan this called radius 1 and ...

[Wolfgang Schwanghart]

Well, then use isnan and brush the data such that only regions of nans remain that are smaller than a specified number of cells (k).

k = 1;

I = isnan(dem);

I = xor(bwareaopen(I,k+1),I);

demcorrected = roifill(dem,I);

I didn't try it since I just have Matlab not at hand, but it should work.

[Wolfgang Schwanghart]

The last line of code should be

demcorrected = roifill(dem,imdilate(I,ones(3)));