Generation of edge stop parameter according to self-snake model

I want to generate an edge stop parameter according to the self snake model by G. Sapiro to remove multiplicative noise. The function is g(r) = 1/(1+(r/k)^2), where r is magnitude of ∇u. u is the actual image. My code is as follows

[GradientX,GradientY] = gradient(double(originimg));
Gr = sqrt((GradientX.*GradientX)+(GradientY.*GradientY));
[X1, Y1]=size(Gr);
for i=1:X1
    for j=1:Y1
        r = mean(Gr(i,:));
    end
end
stParam = 1 / (1 + ((r / gradTP)^2));

But I'm not getting the expected result. Please suggest some way out. Thanks in advance.

[Geoff Hayes]

Rajdeep - please describe the result that you are obtaining. Why is it not the expected one?

[Rajdeep Nandi]

Actually it is giving a matrix as the result, but according to the paper I'm following, the parameter must be a scalar.