Discrete convolution with a non constant kernel

[Laurent Hoeltgen]

Hi,

is there a function in octave that performs a convolution of a 2D array
(i.e. image) with a kernel that is different in each pixel? So basically
in pixel (i,j) I want to compute

sum_(k,l) f(i-k,j-l) * g((i,j), (k,l))

where f is my input image and g my convolution kernel (with dependence
on (i,j)), too. I could implement it by myself, but before I start
hacking something together with too many for loops I wanted to make sure
that there is no function which does the job already. I need such a
function, because I want implement non-linear and anisotropic diffusion
processes.

Regards,
Laurent