Median filter without medfilt2

Hello everyone,

I have to implement the median filter for a gray-scale picture and a variable window(for example 3x3, 5x5, 9x9, etc) without the built in function medfilt2. It's allowed to use median() and sort() and reshape(). In the edges where you cannot build the median, we should output the original value.

Can someone help me?

[Rik]

Question recovered from Google cache:

Hello everyone,

I have to implement the median filter for a gray-scale picture and a variable window(for example 3x3, 5x5, 9x9, etc) without the built in function medfilt2. It's allowed to use median() and sort() and reshape(). In the edges where you cannot build the median, we should output the original value.

Can someone help me?

[Rena Berman]

(Answers Dev) Restored edit

Answer 1

[Matt J]

Since it's a homework assignment, I don't think you need to be computationally eficient. Just loop over all the appropriately sized sub-matrices of the image and apply median() to each... You could alos use blockproc()

https://www.mathworks.com/help/images/ref/blockproc.html

[Franzi]

Yes, I know that it also works with loops, but I wanted to see another version because loops are very inefficient, therefore I wanted to see a more efficient version of how to implement that.

[Matt J]

Well, when efficiency matters, no one would ever do anything else besides medfilt2, or maybe ordfilt2, though I assume that is similarly off-limits to you.

[Franzi]

ok, so there is now way in between medfilt2 and loops?

[Matt J]

None that minimizes memory consumption, I don't believe. You could unfold all of the window data using a very small loop like so,

Image=rand(300); %fake image
N=3; %window size

C=cell(1,N^2);
for i=1:N
    for j=1:N
        C{i,j} = reshape( Image(i:end+i-N,j:end+j-N)  ,[],1);
    end
end

C=cell2mat(C(:).');

Now you have a matrix C whos rows are the different NxN window data. You can now conveniently operate across the rows using sort() or whatever you want.. But note that this consumes N^2 times the amount of memory as a single image:

>> whos C
  Name          Size              Bytes  Class     Attributes

  C         88804x9             6393888  double          

[Franzi]

ok, thank you!

Answer 2

[Image Analyst]

Franzi: I think I already gave you my nlfilter demo in your other question. All you had to do was change the function to do median() instead of thresholding. Here, I've attached a new m-file where I've done this easy change for you. I just replaced about 5 lines of code with a call to median - pretty easy.

[Franzi]

ok, thank you!