We model the SPOT image with a 4-connected Gaussian Markov Random Field .
The associated distribution is given by :
Using the Markov property, the local conditionnal law on the initial lattice is given by a normal law :
Computing the local conditionnal variance on the SPOT data, we get a first estimator :
We next consider the SPOT image on a decimated lattice, keeping one pixel out of two :
We compute the marginal law of the model on the decimated lattice. We obtain an 8-connected Gaussian Markov Random Field .
Using the Markov property, the local conditionnal law on the initial lattice is given by a normal law :
Computing the local conditionnal variance on the decimated SPOT data, we get a second estimator :
Combining the two variances we get a estimator of T :