CODEWITHSUNDEEP
matlabimage-processing
user6934433
user6934433

Reputation:

using convolution function in Matlab

I have a grayscale image. I want to calculate Del for every pixel of image, where 

del =  sum(absoluteValue(anypixelInensity - it's 8 neighbors Pixel Intensity) 

I = imread('crop.bmp');


%KERNEL
windowSize = 3;
kernel = -1 * ones(3,3);
del = abs(conv2(double(I), double(kernel), 'same'));
disp(I)
disp(del)

I want to implement it with matlab conv2 function. Which would subtract current pixel intensity from its neighbors and give the absolute sum.

for example:

10 2 3
12 15 7
4 5 6

del(1,1) = abs(10-2)+abs(10-12)+abs(10-15)
or del(2,2) = abs(15-10)+abs(15-2)+abs(15-3)+abs(15-12)+abs(15-7)+abs(15-4)+abs(15-5)+abs(15-6)

Upvotes: 1

Views: 431

Answers (2)

beaker
beaker

Reputation: 16809

This one is a bit tricky because, as @AnderBiguri says, the abs inside the kernel makes it nonlinear, so conv2 won't work.

You can, however, use nlfilter with a custom function. (This implementation uses nansum, which is in the Statistics Toolbox. If you don't have access to this function, I can give you an alternate function, it's just easier with nansum.)

Another difficulty is that we have to handle the image borders in such a way that they don't mess up the sum. This means that we have handle the image padding ourselves. I've chosen to use nan values to pad. I've also been lazy and assumed an odd value for windowSize. If this is not always the case, we'll have to adjust the padding.

% windowSize assumed to be odd; window assumed to be square
windowSize = 3;

% pad the input image
% padsize is how far the window extends beyond anchor
padsize = floor(windowSize/2);
A = padarray(I, [padsize, padsize], nan);

% create an anonymous function for nlfilter to use
% calculate row/column number of anchor within its neighborhood
anchor = ceil(windowSize/2);
% subtract the anchor element from its neighbors
%   then sum the neighbors *not* equal to NaN
%   (value at anchor position will be 0)
f = @(x) sum(nansum(abs(x-x(anchor,anchor))));

% filter padded image, and then crop off the padding
del = nlfilter(A, [windowSize, windowSize], f);
del = del(1+padsize:end-padsize, 1+padsize:end-padsize);

Upvotes: 1

Ander Biguri
Ander Biguri

Reputation: 35525

Yes, conv2 function does the convolution itself.

Given an image, A and a kernel k

imout=conv2(A,k);

Will give you the convolution of the kernel over the image.

Read the documentation

However, you may want to use imfilter, as it s upports better different types of images.

Upvotes: 0

Related Questions