Faster Image Processing?

Hello,

I'm trying to make my code faster and I haven't been successful and I'm hoping that someone has great ideas.

I have 3D arrays: Image(512,512,60) and RefImage(1000,288,16).

I'm trying to update the 3D Image array based on the calculations using the values in RefImage pages(16 of them).

Here's the simplified version of what I'm trying to make it faster:

for i = 1:1000
    for j = 1:288
        RCPointer(varies depends on how many row and col are occupied) = pointer{j,i};
        ImagePage(16 of them) = page{j,i} * ConstantMultiplier;
        for k = 1:16
            ImagePointer = (ImagePage(k)-1) * 262144 + RCPointer;
            Pixel = Image(ImagePointer);
            NewPixel = Pixel + RefImage(i,j,k);
            Image(ImagePointer) = NewPixel;
        end
    end
end

Thanks for your ideas!

[Milan Bansal]

It is difficult to know what this code is trying to do from the simplified version. It will be better if you can provide the full code.

[DB]

Sorry, I can't provide the full code. However, the main purpose should be be pretty clear from the simplified code above. I need to update the Image array(512x512x60) with the RefImage(1000x288x16) array values. Where to update is based on the cell array pointer and page values.

[Catalytic]

However, the main purpose should be be pretty clear from the simplified code above.

No, because code isn't enough to tell us what's going on. We need to see input data as well, so we can understand the size and shape of the matrix variables it contains. If you're going to provide simplified code, you need to provide simplified input data to go with it - in this case pointer, page, and ConstantMultiplier - and we should be able to run the code with the input you provide.

Answer 1

[Constantino Carlos Reyes-Aldasoro]

Two nested for loops with images is not an efficient way to deal with matrices or images. The point of Matlab is to handle everything as a matrix. I.e.

a=ones(5,5);
b=2*a;

is efficient as opposed to

for k1=1:5
    for k2=1:5
        b(k1,k2) = a(k1,k2) *2;
    end
end

Now, how do you handle the to and from matrices? You can use indices, i.e., addressing the matrix. For example

a= ones(5,5);

a = 5×5

1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1

a(2:4,2:4) =2;
b= 2*a;
b(a==2) = 5;
disp (b)

     2     2     2     2     2
     2     5     5     5     2
     2     5     5     5     2
     2     5     5     5     2
     2     2     2     2     2

In that way, you do not need to use for loops, and can directly operate in certain regions of the images/matrices (i.e. the address that you pass, in this case a==2)

I wrote a book on image analysis, addressing a matrix is part of chapter 1:

https://onlinelibrary.wiley.com/doi/epdf/10.1002/9781118657546.ch1

Answer 2

[Matt J]

[m,n,p]=size(Image);

N=cellfun(@numel, pointer(:));


K=cell2mat( cellfun(@(q)q(:)', page(:),'uni',0)); %*ConstantMultiplier? No idea what that was doing
K=repelem(pointer,N,1);


[I,J]=ind2sub([m,n], vertcat(pointer{:}));
I=repmat(I,1,width(K));
J=repmat(J,1,width(K));

Update=accumarray([I(:),J(:),K(:)],  RefImage(:), [m,n,p]);

Image = Image + Update;