CODEWITHSUNDEEP
matlabsignalssignal-processinginterpolationvectorization
nw.
nw.

Reputation: 5155

Vectorized Pixel Doubling Interpolation in Matlab

I have an NxMx3 matrix of unit8s, with each (x,y,:) being a vector of RGB values. I'm trying to do pixel doubling interpolation by simply copying the east and south neighboring values over by one pixel. 

eastNeighbors = [img(2:end,:,:)];
southNeighbors = [img(:,2:end,:)];

filtered = img + eastNeighbors + southNeighbors;

The problem with this is that I get a "Array dimensions must match for binary array op." error.

How can I add a row and column of [0, 0, 0] elements to eastNeighbors and southNeighbors, respectively?

Edit: Actually, shouldn't there be a way to implement this with filter2? This is basically the FIR filter out(x,y) = in(x,y) + in(x-1, y-1), correct?

Tried this:

h = [[1 1] [1 1]];

filtered = filter2(h, img);

Got "Not enough input arguments." error.

Upvotes: 2

Views: 156

Answers (3)

Divakar
Divakar

Reputation: 221574

This might be what you are after -

Approach 1: Using nlfilter for sliding-neighborhood operations

fun1 = @(x) sum([x(2,2) x(3,2) x(2,3)]);
OUTPUT_IMG = nlfilter(INPUT_IMG, [3 3], fun1);%%// INPUT_IMG is a gray image

Approach 2: Using direct summations by padding zeros with padarray to match matrix sizes

southNeighbors = padarray(INPUT_IMG(2:end,1:end),[1 0],'post');
eastNeighbors = padarray(INPUT_IMG(1:end,2:end),[0 1],'post');
OUTPUT_IMG = INPUT_IMG + eastNeighbors + southNeighbors;

General Notes:

If you are looking for interpolation, maybe you would want to find mean of those three pixels, so divide by 3. Take care of the uint8 datatype as it will clip off values at 255 when summing up.

Also, you might want to keep the last row and column same as in the original image, as they don't have both east and south neighboring pixels, so use this -

OUTPUT_IMG(:,end) = INPUT_IMG(:,end); 
OUTPUT_IMG(end,:) = INPUT_IMG(end,:);

Upvotes: 3

Rafael Monteiro
Rafael Monteiro

Reputation: 4549

You could do it with convn, like this:

filtered = convn(img, [0 1 0; 1 1 0; 0 0 0]);

Upvotes: 1

Brian
Brian

Reputation: 617

To pad the images like you are interested in, you can use a combination of cat() and zeros() like so:

eastNeighbors = cat(2, zeros(size(img,1), 1, size(img,3)), img(:,2:end,:));
southNeighbors = cat(1, zeros(1, size(img,2), size(img,3)), img(2:end,:,:));

filtered = img + eastNeighbors + southNeighbors;

Upvotes: 1

Related Questions