Semantic Visualization Facilitated By Cluster Analysis

1
Semantic Visualization Facilitated By Cluster
Analysis

• Eun Ju Nam, Mauricio Mauer,
• and Klaus Mueller
• Center for Visual Computing,
• Stony Brook University

2
Volume Classification

• The density and gradient values in volumetric
  data most often used in transfer function guided
  feature detection and enhancement.
• However, there are more useful metrics to reveal
  additional information.
• We derive more parameters/signatures than the
  density by transformation.
• Signatures
• Density
• Up-history / Down-history (LH values)
• Gradient magnitude
• Second order moment
• Third order moment
• Kurtosis
• Skew
• Curvature
• Texture
• and so on.

3
Cluster Analysis and Learning

• Introduction of a large number of metrics creates
  high-dimensional classification space.
• It is impossible to handle with the transfer
  function interface.
• We propose Cluster Analysis for this task.
• Cluster Analysis
• A strategy to explore high dimensional data
  without a classification model or a
  priori-hypothesis.
• Find a collection of features to determine an
  object using CA.
• Learn from classification models which are formed
  by CA.

4
Semantic Query

• Semantic query for a specific object
• Display the boundary voxels of significant
  objects with density D, each such object in a
  different color.
• Transform to the specific feature space
• Display only voxels with a high gradient G and
  an H value of density D, but with any L value and
  separate all the connected regions with more than
  X voxels.
• Sequential query processing at the given query.
• Clustering on gradient magnitude to identify
  candidates of boundary voxels.
• Clustering on up history (the H-value).
• Clustering on intensity to distinguish the
  various boundaries.
• Clustering on spatial connectivity (position) to
  separate regions.
• Eliminate clusters which have less than X voxels.

5
Cluster Analysis To find out Features
Semantic Query
Display
Learning Objects Associated with rules
model
Transform the query to Signature space
Classify voxels With signature queries
6
Related Works

• Identify region boundaries using 2D
  density-gradient plot G. Kindlmann, J. W.
  Durkin 98
• Multidimensional Transfer Function using
  dual-domain J. Kniss, G. Kindlmann, C. Hansen
  02
• Painting-based High D Classification F. Tzeng,
  E. B. Lum, K. Ma 05
• Visualization Boundaries using LH Histograms P.
  Sereda, A. V. Bartoli, I. Serlie, F. Gerritsen
  06
• Query-Driven Visualization of Large Data Sets
  K. Stockinger, J. Shalf, K. Wu, E. W. Bethel 05
• Vector Field Hierarchies B. Heckel, G. Weber,
  B. Hamann, K.I. Joy 99

7
Initial Results of CA with Cello Volume Data
8
Volume Signature Space
9
Slice-based Spatial Space
10
Clustering on gradient magnitude to
identify candidates of boundary voxels.
11
Clustering on up history (the H-value).
12
Clustering on intensity to distinguish the D
intensity boundaries.
13
Clustering on spatial connectivity(position) to
separate regions.
14
Conclusion

• We outlined the beginnings of a framework that
  generates, capture, and encodes knowledge derived
  from the analysis of volume dataset.
• Users simply formulate a query. The system
  performs the required voxel clustering based on
  derived knowledge.
• The knowledge is generated by a dual-domain
  interactive clustering framework.
• We demonstrated the framework using a relatively
  simple cello dataset.

15
Acknowledgements

• This work was supported by the U.S. Department of
  Energy Office of Basic Energy Sciences, Chemical
  Sciences Division.

16
Question?!Thank you!