Store a group of points (given x, y, and value) into image efficiently

I have a matrix M with size m x 3 , where m is the total number of points (thousands of points). The three columns correspond to x, y and value. I want to map the values into an image efficiently. This task is very easy, but I need to speed up the process. I came up with the following two approaches:

First approach (Testing on 3600 points, I got a run time of 0.07 seconds.)

a = M(:,2);
b = M(:,1);
c = M(:,3);
img= accumarray ([a(:),b(:)],c(:));
img(1000,1) = 0;
img(1,1000) = 0;

Second approach (Testing on 3600 points, I got a run time of 0.04 seconds.)

img = zeros(1000,1000);
for i = 1 : size(M,1)
    img(M(i,2),M(i,1)) = M(i,3);
end

parfor is not an option here in the second approach.

Does anyone know how to make this process faster?

Please note that I excute this code 50 times per frame obtained from a camera, resulting with a 2 second run time per frame.

[Yousef Atoum]

Can anyone make this task faster?

[Guillaume]

As I said, it's unlikely you'll find anything faster. Your loop code and my answer do the bare minimum: memory allocation (with zeros) then simple memory copy using an offset. At this point, the only thing that could make it faster is if the data to be copied is ordered with regards to the destination.

Answer 1

[Guillaume]

img = zeros(1000, 1000);
img(M(:, 2) + 1000 * (M(:, 1) - 1)) = M(:, 3);

[Yousef Atoum]

Thanks Guillaume! I like your answer!

However, It did not improve the efficency. It gave me the same run speed as my for loop in my second approach (with a slight average speed up of 0.001).

[Guillaume]

Other than resorting to mex, I don't see how you can get any faster. Even with a mex, I'm not convinced you'll get any improvement.