A question about inpaint_nans.m

I have worked with this great function and I can't understand why this happens: If we have a vector with 100 data and some nans and run the function to get the values for nans and also dividing the data to 5 parts(5*20 each small vector has 20 data) and using the function(the nans are the same for each 20 records ) so why the results for both ways are not the same??? Thanks for any idea.

Answer 1

[John D'Errico]

Why would they be the same? You are posing two different problems. It sounds like you have reshaped the vector into a 5x20 array. On two dimensional problems, the function works in both dimensions, looking at the neighbors of a point in all directions to interpolate.

Or, did you break the vector into 5 disjoint vectors, then calling inpaint_nans 5 separate times? Even in this case, the function uses the neighbors of a point in each direction. So if a nan is at or near the break point between those segments then of course there will be a difference.

Perhaps you can give an example where you see a problem, attach the data to a comment, and show how you called it.

[Rita]

Thanks John,the data is in 1D shape.and Yes I took a part the vector to 5 "disjoint vectors" The run the function for 5times separately.But I have not noticed that I have nans at the breakpoints! Thanks a lot for the clarification.

[John D'Errico]

Unless you can give an example where you see something different, it is difficult to know what you saw.

V = 1:20;
>> V([5 14]) = NaN;
>> inpaint_nans(V)
ans =
  Columns 1 through 12
            1            2            3            4            5            6            7            8            9           10           11           12
  Columns 13 through 20
           13           14           15           16           17           18           19           20
>> V1 = V(1:10);
>> V2 = V(11:20);
>> inpaint_nans(V1)
ans =
            1            2            3            4            5            6            7            8            9           10
>> inpaint_nans(V2)
ans =
           11           12           13           14           15           16           17           18           19           20

As you can see, both cases produce identical results for the NaN elements.

[Rita]

Thanks John

Answer 2

[Image Analyst]

FYI, starting with R2015a there is a built-in function regionfill().

[Rita]

Thanks Image Analyst for the info.I used like "fillgaps" function but it seems that finding the optimal order and maximum length was a bit bother.

[Image Analyst]

Actually I was wrong - it can't be used unless you supply a 2-D image. It doesn't work with 1-D vectors.

[John D'Errico]

I had to modify inpaint_nans to work on vectors. It uses the same idea, but only looks along the length of the vector.

[Rita]

But still inpaint_nans works well for the vectors. I have used the function for the vectors...with model number 4(metaphor)Thanks