Statistical Image Modeling and Processing Using Wavelet-Domain Hidden Markov Models - Nonlinear Signal and Image Processing: Theory, Methods and Applications - page 345

1

0.8

0.6

0.4

0.2

1:Using the proposed scheme
2:Using the neutral setting

0

0 10 20 30 40 50 60 70 80 90 100

Iteration Number of EM Training

(a)

1

0.8

0.6

0.4

0.2

1:Using the proposed scheme
2:Using the neutral setting

0

0 10 20 30 40 50 60 70 80 90 100

Iteration Number of EM Training

(b)

FIGURE 10.6
The plots of ln f

2
η =

0
are tested, where the proposed initialization scheme (solid line: 1) and the neutral initial setting
(dotted line: 2) are used, respectively. (a) Doppler (1: MSE

( w | θ) with respect to the EM iteration number. Four noisy signals σ 1 .

= 0.111, 2:MSE = 0.163). (b) Bumps

(1: MSE = 0.306, 2: MSE = 0.532). (b) Blocks (1: MSE = 0.256, 2: MSE = 0.468). (d) Heavisine

(1: MSE = 0.070, 2: MSE = 0.095). ( Continued )