Previous Contents Next
Previous : Reconstruction algorithm  Next : Reconstruction results 
Estimation results
Set of hyperparameters which cancels
the log-likelihood derivatives
(local minima)
see pictures below
Original image 
64 x 64
  $\lambda $=0.13$\delta $=2 
SNR=27.5 dB
$\lambda $=0.25$\delta $=5 
SNR=28.9 dB
$\lambda $=0.35$\delta $=10 
SNR=28.4 dB
Observed image 
SNR=15.9 dB
  $\lambda $=0.66$\delta $=30 
SNR=26.3 dB
$\lambda $=0.91$\delta $=50 
SNR=25.0 dB
Original image 
"plant" 128 x 128
  $\lambda $=0.19$\delta $=2 
SNR=22.70 dB
$\lambda $=0.33$\delta $=5 
SNR=22.72 dB
$\lambda $=0.49$\delta $=10 
Observed image 
  $\lambda $=1.2$\delta $=30 
SNR=22.79 dB
$\lambda $=1.8$\delta $=50 
SNR=22.73 dB

 Comparison between different reconstruction results
corresponding to a set of optimal hyperparameters
Signal to Noise Ratio (SNR) - "plant" image
around the optimal value of =0.19 (=2)
The best SNR really corresponds to the estimated.
Previous Next
André Jalobeanu - 24 Aug 1998