Digital Signal Processing Reference
In-Depth Information
Fig. 13.9
Hard thresholding
and soft thresholding
Donoho and Johnstone [
13
] have shown that the choice of
S
is near-optimal for
N
4.
Fig.
13.9
represents the thresholding procedure. Note that hard thresholding cre-
ates discontinuities at
k
≥
1.
Heuristic denoising is based on the hard and soft thresholding functions (see
Fig.
13.9
), the non-negative garrote (NNG) function and the smoothly clipped abso-
lute deviation (SCAD) function, both illustrated in Fig.
13.10
.
Although these standard thresholding functions are close to optimal, they raise
some limitations:
±
•
The hard thresholding function is not everywhere continuous and its discontinu-
ities at
k
±
1 generate a high variance in the estimated signal;
Fig. 13.10
NNG and SCAD shrinkage functions
Search WWH ::
Custom Search