Image Denoising Using Wavelets: Application in Medical Imaging - Advances in Heuristic Signal Processing and Applications - page 298

Fig. 13.14 Denoising based on estimation of the probability of the presence of useful information:
3-band Lena (RGB)
In this context, only advanced heuristics are suitable to properly denoise images
corrupted by correlated noise. These methods combine parametric methods based
on a priori knowledge (or statistical modeling) and nonparametric methods such as
thresholding in the wavelet domain. Among the methods which demonstrate some
effectiveness is the work of A. Pizurica [ 17 ]. The heart of this approach is to esti-
mate the probability that a given wavelet coefficient contains a significant noise-free
component. Heuristically, the signal of interest is identified and extracted from the
noisy image. Figure 13.14 shows the application of this method to a three band
image (RGB), specifically to the color Lena image.
A few years later, another method was proposed by the same team [ 18 ] to extract
the useful signal. This is characterized by measuring and quantifying the relevant
information in a noisy image taking into account the structure of the correlation of
neighboring wavelet coefficients. The approach consists of combining an intra-scale
model with a hidden Markov type model to capture these dependencies between
wavelet coefficients.
Figure 13.15 shows an example of application of the denoising method based on
a Markov model and a slightly redundant discrete wavelet transform. This method is
compared with SNR: it appears that the denoising presented in Fig. 13.15 (d) offers
a slight improvement.
13.4 Applications in Medical Imaging
In recent years, medical imagery and diagnostic techniques have seen spectacular
developments and heavy investment, and research hospitals in particular have es-
tablished neuroimaging centers equipped with functional PET (positron emission
tomography) and MRI (magnetic resonance imagery) equipment.
These new technologies supplement the more classic techniques, which are also
perpetually evolving, for example, ultrasound, X-ray tomography (or classical scan-
ning), magnetoencephalography (or MEG, offering access to the spatiotemporal