Hi all!

I'm trying to realize the work of Sadjadi named "A full plane block
kalman filter for image restoration". In this paper, there is a
section to find a parameter named "p".

Simply, I can explain my problem like this. There are some vector
observations, x, that represents some pixel values on the image and
their sizes are 1xN. It is wanted to find the parameter p(k) and its
definition is given like:

p(k) = E[x(n)transpose(x(n-k))] where n = 1:M (M observations)

I will try to get this parameter by finding covariance matrix and adding
the mean(X) to it.

Because, cov(X(n),X(n-k)) = E[x(n)transpose(x(n-k))] - mean(X)^2;

But, i have a problem at this point. For example,

Define X = [x(1);x(2);...x(M)];
The size of X : MxN (M observations and N variables)

- If cov(X(n)) is called, it returns NxN matrix as i expected.

- But, if cov(X(n),X(n-k)) is called, it returns 2x2 matrix.

My question is that,
"Why there occurs such a dimension difference?" Start a New Thread