Can someone please help me implementing this code without loop ?

Hello,

Can someone help me to speed up this code? Without loops

img = imread("my_img.png");
for i=2:256
    for j=2:256
        v1(i,j) = max(abs(img(i+1,j)-img(i,j)),abs(img(i-1,j)-img(i,j)));
        v2(i,j) = max(abs(img(i,j-1)-img(i,j)),abs(img(i,j+1)-img(i,j)));
        v3(i,j) = max(abs(img(i-1,j+1)-img(i,j)),abs(img(i+1,j-1)-img(i,j)));
        v4(i,j) = max(abs(img(i+1,j+1)-img(i,j)),abs(img(i-1,j-1)-img(i,j)));
    end
end
min_v1 = min(v1);
min_v2 = min(v2);
min_v3 = min(v3);
min_v4 = min(v4);

Thanks

[Bruno Luong]

Since your loop does store intermediate results, you don't need loop at all, just do the last one:

img = imread("my_img.png");
i=256
j=256
v1=max(abs(img(i+1,j)-img(i,j)),abs(img(i-1,j)-img(i,j)));
v2=max(abs(img(i,j-1)-img(i,j)),abs(img(i,j+1)-img(i,j)));
v3=max(abs(img(i-1,j+1)-img(i,j)),abs(img(i+1,j-1)-img(i,j)));
v4=max(abs(img(i+1,j+1)-img(i,j)),abs(img(i-1,j-1)-img(i,j)));
[x,y]=min([v1,v2,v3,v4]);

[Bruno Luong]

I think your code is buggy. Solve the bug then you can ask for optimization.

[Rik]

It looks like you are trying to implement something to do with a GLCM. You can check either the FEX or the native function to see if it is something you can use.

[pb1106]

I'm trying to implement a form of the Interest Point Operator in the V8 space for a pixel , given the equation that I wrote before.

[Bruno Luong]

Hate to repeat myself, now you correct v, but this looks buggy still:

[x,y]=min([v1,v2,v3,v4])

[pb1106]

@Bruno Luong, I've edited in the correct form now

[Bruno Luong]

Repeat myself a third time. Here is how your edited code works:

img = rand(4,5)
[m,n] = size(img);
for i=2:m-1
    for j=2:n-1
        v1(i,j) = max(abs(img(i+1,j)-img(i,j)),abs(img(i-1,j)-img(i,j)));
        v2(i,j) = max(abs(img(i,j-1)-img(i,j)),abs(img(i,j+1)-img(i,j)));
        v3(i,j) = max(abs(img(i-1,j+1)-img(i,j)),abs(img(i+1,j-1)-img(i,j)));
        v4(i,j) = max(abs(img(i+1,j+1)-img(i,j)),abs(img(i-1,j-1)-img(i,j)));
    end
end
min_v1 = min(v1)
min_v2 = min(v2)
min_v3 = min(v3)
min_v4 = min(v4)

img =

    0.1544    0.8711    0.5306    0.2992    0.5583
    0.3813    0.3508    0.8324    0.4526    0.7425
    0.1611    0.6855    0.5975    0.4226    0.4243
    0.7581    0.2941    0.3353    0.3596    0.4294

min_v1 =

     0     0     0     0

min_v2 =

     0     0     0     0

min_v3 =

     0     0     0     0

min_v4 =

     0     0     0     0

>>

The optimization code is very simple (no need for-loop at all)

min_v1 = zeros(1,n-1)
min_v2 = zeros(1,n-1)
min_v3 = zeros(1,n-1)
min_v4 = zeros(1,n-1)

[pb1106]

Ok, I understand that I forgot the pre-allocation but what is the connection between pre-allocation and no-loop code ?

[Bruno Luong]

Preallocation or not it's still give outcome vectors containing only 0s.

At this point I'm not even sure the formula inside the for-loop is really what you want, since there are still errors somewhere else after 2/3 corrections.

I suggest you debug your code and post the version that works as you expected, then ask the question because we can't help you to remove a loop if we don't know what is the expected result.

[pb1106]

This is what I'm trying to implement

Answer 1

[Bruno Luong]

f = rand(4,5);

[m,n] = size(f);
x = 2:m-1;
y = 2:n-1;
v1 = max(abs(f(x+1,y)-f(x,y)),abs(f(x-1,y)-f(x,y)));
v2 = max(abs(f(x,y-1)-f(x,y)),abs(f(x,y+1)-f(x,y)));
v3 = max(abs(f(x-1,y+1)-f(x,y)),abs(f(x+1,y-1)-f(x,y)));
v4 = max(abs(f(x+1,y+1)-f(x,y)),abs(f(x-1,y-1)-f(x,y)));
vmin = min(cat(3,v1,v2,v3,v4),[],3);
alpha = 0.1;
[x,y] = find(vmin>alpha);
x = x+1;
y = y+1;

I_f = [x y]