Image Processing Reference
In-Depth Information
on areas of high uncertainty. In a similar fashion, contours already present in the
visualization can be used [ 84 , 85 ] or modified [ 71 , 92 ] to express uncertainty.
Because not all data is visualized effectively using glyphs, the addition of glyphs
to convey only uncertainty information is often a preferable approach. A specific
example is the UISURF system [ 45 ], which visually compares isosurfaces and the
algorithms used to generate them. In this system, glyphs are used to express positional
and volumetric differences between isosurfaces by encoding the magnitude of the
differences in the size of the glyphs. Similarly, line, arrow, and ellipsoidal glyphs
can be used to depict uncertainty in radiosity solutions, interpolation schemes, vector
fields, flowsolvers, astrophysical data and animations through variation of placement,
magnitude, radii, and orientation [ 54 , 55 , 57 , 75 , 91 , 93 , 109 , 110 , 113 ].
1.5.2.4 Image Discontinuity
Uncertainty visualization often relies on the human visual systems ability to quickly
pick up an images discontinuities and to interpret these discontinuities as areas with
distinct data characteristics. Techniques that utilize discontinuities rely on surface
roughness, blurring, oscillations [ 13 , 33 , 56 , 108 ], depth shaded holes, noise, and
texture [ 22 ], as well as on the translation, scaling, rotation, warping, and distortion of
geometry already used to visualize the data [ 75 ], to visualize uncertainty. Animation
can highlight the regions of distortion or blur or highlight differences in visualization
parameters [ 30 , 60 , 66 ]. Such techniques have been applied to multivaritate data
displayed through scatter plots or parallel coordinates [ 27 , 36 ].
1.6 Examples
1.6.1 Medical Visualization
A fundamental task in medical visualization is segmentation, the partitioning of a
given image into regions that correspond to different materials, to different anatomi-
cal structures, or to tumors and other pathologies.Medical image acquisition typically
introduces noise and artifacts, and we may wish to segment structures for which the
data itself provides little contrast. This is a source of data uncertainty. In many cases,
segmentation also involves complex computational models and numerous parame-
ters, which introduces model uncertainty.
Traditional volume rendering classifies materials based on scalar intensity or fea-
ture vectors that account for first and second derivatives [ 50 ]. Lundström et al. [ 60 ]
introduce probabilistic transfer functions that assign material probabilities to model
cases in which the feature ranges of different materials overlap. This results in a dis-
tribution of materials at each location in space, which is visualized by an animation
in which each material is shown for a duration that is proportional to its probability.
 
Search WWH ::




Custom Search