Biomedical Engineering Reference
In-Depth Information
decomposition of the image information provides the possibility of analyzing
the coarse resolution first, and then sequentially refines the segmentation result
at more detailed scales. In general, such practice provides additional robustness
to noise and local maxima.
In [100], image data was first decomposed into “channels” for a selected set
of resolution levels using a wavelet packets transform. An MRF segmentation
was then applied to the subbands coefficients for each scale, starting with the
coarsest level and propagating the segmentation result from one level to initialize
the segmentation at the next level.
More recently, Davatzikos
et al.
[101] proposed hierarchical active shape
models where the statistical properties of the wavelet transform of a deformable
contour were analyzed via principal component analysis and used as priors for
constraining the contour deformations.
Many research works beneficially used image features within a spatial-
frequency domain after wavelet transform to assist the segmentation. In [102]
Strickland
et al.
used image features extracted in the wavelet transform do-
main for detection of microcalcifications in mammograms using a matching
process and
a priori
knowledge on the target objects (microcalcification). In
[103], Zhang
et al.
used a Bayes classifier on wavelet coefficients to determine
an appropriate scale and threshold that can separate segmentation targets from
other features.
6.5 Image Registration Using Wavelets
In this section, we give a brief overview of another very important application
of wavelets in image processing: image registration. Readers interested in this
topic are encouraged to read the references listed in the context.
Image registration is required for many image processing applications. In
medical imaging, co-registration problems are important for many clinical tasks:
1. multimodalities study,
2. cross-subject normalization and template/atlas analysis,
3. patient monitoring over time with tracking of the pathological evolution
for the same patient and the same modality.
