Digital Signal Processing Reference
In-Depth Information
Fig. 14.8
Separation results of in vivo prostate 1H MRS spectra: (
a
) an observed spectrum from
cancerous tissue (
solid line
) and the estimated baseline (
dashed line
); (
b
) an observed spectrum
from normal tissue (
solid line
) and the estimated baseline (
dashed line
); (
c
) the estimated reso-
nances with the spectrum in (
a
); (
d
) the estimated resonances with the spectrum in (
b
)
observed signal into different dictionaries, which can only sparsely represent one of
the source signals. The a priori knowledge about the features of source signals can
be used to construct these dictionaries. For example, in the application to analyze
MRS data, the mathematical model of source signals (resonances of interest) and
the range of model parameters are exploited for the dictionary construction. As for
the procedure of sparse decomposition, many pursuit algorithms in the literature
are available. However, these algorithms could have different performances on a
specific application. Furthermore, researchers sometimes should develop existing
algorithms according to the specific application for achieving a satisfying separation
result. Additional constraints could lead to a better separation performance.
References
1. Aharon, M., Elad, M., Bruckstein, A.: K-SVD: an algorithm for designing overcomplete dic-
tionaries for sparse representation. IEEE Trans. Signal Process.
54
(11), 4311-4322 (2006)
Search WWH ::
Custom Search