Can I use CVX for images (in a reasonable amount of time)?

1

Hi all,

I’d like to use the following code on my images (A(x) is an operation, x is the image, and b is the measured data).

cvx_begin
    variable x(N,N))
    minimize( (norm(x,1)) )
    subject to
      norm(A((1-x))-b) <= epsilon
cvx_end

When the images are small 32x32 this is pretty quick but for real images it seems to not be a good solver. Even 64x64 is slow. Can I use CVX for images (in a reasonable amount of time)?

Thanks so much for your help!

Best,

Christine

2

CVX is for convenience, not for speed, and the algorithms it uses are chosen for accuracy over speed as well. Your choice of norm in the constraint necessitates the use of a semidefinite cone, which is definitely going to be slow.

3

I figured. I’ll test with cvx and then move to a more specialized solver.

Thanks for your help!

Christine

4

I figured. I’ll test with cvx and then move to a more specialized solver.

Thanks for your help!

Christine