A code faster than the speed of light ...

Well, may be it's an exaggeration, but now that the speed of light can be exceeded (may be) we might as well use that phrase.

So we had fun with the Sobel filter yesterday. But remember we used nested for loops. If you are serious about faster computations, you should always avoid the for loops. Why not use the magical command of circshift instead in coding the sobel filter?

Here is a comparison of the performance of Sobel filtering with and without using the for loops.
>> u=imread('barbara.png'); t1=cputime; [A1 ux1, uy1]=sobel_for(u); t1=cputime-t1


t1 =


    5.0500


>> u=imread('barbara.png'); t=cputime; [A ux, uy]=sobel(u); t=cputime-t


t =


    0.0700


>> t1/t


ans =


   72.1429

This is very nice the new code is 72 times faster that the previous code. Me likey!

Smart display of images

The command likes the (double) images if they are scaled from 0 to 1.
So we do it as follows:
m=min(min(min(A))); M=max(max(max(A))); figure; imshow((A-m)/(M-m));
m=min(min(min(ux))); M=max(max(max(ux))); figure; imshow((ux-m)/(M-m));
m=min(min(min(uy))); M=max(max(max(uy))); figure; imshow((uy-m)/(M-m));
Here is the output of the images:

Output of the Sobel filter

The x-component of the Sobel filter

The y-component of the Sobel filter