Image Denoising Using Wavelets: Application in Medical Imaging - Advances in Heuristic Signal Processing and Applications - page 304

Fig. 13.18 Left : original MR
image. Right : Image noised
by Rician noise ( σ = 30,

SNR = 5 . 9dB)

Fig. 13.19 Left :Image
denoised by a
spatially-adaptive Wiener
filter (SNR 10 . 1dB).

Right : Image denoised by the
algorithm proposed for K = 2
and a window of analysis of
size 3 × 3(SNR = 12 . 9dB)

=

Performances of the proposed method are illustrated on an MR image with ar-
tificially added Rician noise (see Fig. 13.18 ), and compared to spatially-adaptive
Wiener filtering. Figure 13.19 shows that, qualitatively, denoising using the pro-
posed method clearly surpasses spatially-adaptive Wiener filtering. MR images were
provided by the Gand (Belgium) university hospital. Suppression of noise in these
images facilitates subsequent automatic processing such as segmentation, for exam-
ple.
The optimal level of wavelet decomposition is J

=

4. The tuning parameter
value K , which appears optimal, is 3 in echography (see Fig. 13.21 )and2inMRI
(see Fig. 13.19 ). An increment in this parameter can smooth the image, and con-
sequently can lead to a loss of information. The spatial activity indicator e(k) is
calculated by locally averaging neighboring coefficients. A 3

× 3 window is inter-

esting in terms of SNR in echography as well as in MRI.
This method, proposed by Aleksandra Pizurica [ 17 ] is of low complexity, both
in terms of implementation and execution time, and adapts to unknown noise and to
the local context of the image. The results produced have been demonstrably useful