Previous : Sampling the densities Next : The MCMCML estimation algorithm |
|
auxiliary variable b, auxiliary images Bx, By (gradients w.r.t. x and y)
sampling triplets (X, Bx, By) in an alternated way on X and Bx, By
the pixels of Bx and By are independent
and the law of X knowing Bx, By is Gaussian
sampling X is made in a single pass (unlike Gibbs and Metropolis samplers),
the processing is independent of the neighbourhood size
which provides a fast algorithm for posterior distribution sampling.
|
![]() We first calculate the Fourier transform F[h4 ] and DCT[Y], and W : |
![]() The pixels b of Bx and By are independent : |
![]() The distribution of X with known B is Gaussian because |
|
![]() |
![]() |
![]() |
![]() |
|
|
![]() ![]() |
|
![]() |
![]() |
|
|
![]() |
|
![]() |
![]() |
|
|
![]() ![]() |
|
![]() |
![]() |
|
|
![]() |
|
|
![]() |
|
Prior distribution
|
![]() We first calculate the Fourier transform F[h4 ] and W0 : |
![]() The pixels b of Bx and By are independent : |
![]() The distribution of X with a known B is Gaussian because |
|
![]() |
|
![]() |
![]() |
|
|
![]() ![]() |
|
![]() |
![]() |
|
|
![]() |
|
![]() |
![]() |
|
|
![]() ![]() |
|
![]() |
![]() |
|
|
![]() |
|
|
![]() |
|