Biomedical Engineering Reference
In-Depth Information
improve image quality for PET images that were usually degraded by low reso-
lution and high level of noise.
6.4 Image Segmentation Using Wavelets
6.4.1 Multiscale Texture Classification
and Segmentation
Texture is an important characteristic for analyzing many types of images, in-
cluding natural scenes and medical images. With the unique property of spatial-
frequency localization, wavelet functions provide an ideal representation for
texture analysis. Experimental evidence on human and mammalian vision sup-
port the notion of spatial-frequency analysis that maximizes a simultaneous
localization of energy in both spatial and frequency domains [69-71]. These
psychophysical and physiological findings lead to several research works on
texture-based segmentation methods based on multiscale analysis.
Gabor transform, as suggested by the uncertainty principle, provides an op-
timal joint resolution in the space-frequency domain. Many early works utilized
Gabor transforms for texture characteristics. In [27] an example is given on
the use of Gabor coefficient spectral signatures [72] to separate distinct textu-
ral regions characterized by different orientations and predominant anisotropic
texture moments. Porat et al. proposed in [28] six features derived from Gabor
coefficients to characterize a local texture component in an image: the dominant
localized frequency; the second moment (variance) of the localized frequency;
center of gravity; variance of local orientation; local mean intensity; and vari-
ance of the intensity level. A simple minimum-distance classifier was used to
classify individual textured regions within a single image using these features.
Many wavelet-based texture segmentation methods had been investigated
thereafter. Most of these methods follow a three-step procedure: multiscale
expansion, feature characterization, and classification. As such, they are usually
different from each other in these aspects.
Various multiscale representations have been used for texture analysis.
Unser [73] used a redundant wavelet frame. Laine et al. [74] investigated a
wavelet packets representation and extended their research to a redundant
wavelet packets frame with Lemari e-Battle filters in [75]. Modulated wavelets
Search WWH ::




Custom Search