Interactive multiscale exploration for volume classification

1
Interactive multi-scale exploration for volume
classification

• E.B. Lum, J. Shearer, K.L. Ma
• Pacific Graphics, 2006
• Presenter Kuang-Wei Fu
• Spring. 18th, 2007

2
Outline

• Introduction
• Filter bank
• User interface
• Result Discussion
• Conclusion

3
Introduction

• Transfer function
• Map voxel values to optical properties
• The very act of specifying a transfer function
  promotes data understanding
• Two criteria to choose transfer function
• User ability
• Segmentation methods
• To take local texture and neighborhood
  characteristics into account
• Filter bank

4
IntroductionMain contribution

• Identification of a set of filters that map well
  to the capabilities of current graphics hardware
• The design of an intuitive user interface for
  effectively exploring and utilizing the
  multi-scale volume data

5
IntroductionSystem hierarchy
6
Outline

• Introduction
• Filter bank
• User interface
• Result Discussion
• Conclusion

7
Filter bankDefinition

• Consists of a series of filters that splits a
  signal into a number of sub-bands for subsequent
  analysis
• Each sub-bands provides a different estimate of
  the local energy distribution
• Texture classification in the field of Image
  Processing

8
Filter bankDiagram
9
Filter bankFilter choosing task

• Limited amount of texture memory leads two
  question
• Applying N-sized filter in a pre-processing step
  would use a great deal of texture memory
• Applying on-the-fly would impact interactive
  performance

10
Filter bankGaussian filters

• A sequence of Gaussian low-pass filter can
  efficiently stored in an oct-tree
• Supported in current graphics hardware with
  mip-mapped textures
• Known Gaussian pyramid in the field of Image
  Processing

11
Filter bankGaussian filters diagram
12
Outline

• Introduction
• Filter bank
• User interface
• Result Discussion
• Conclusion

13
User interfaceParallel coordinates
14
User interfaceMaterial percentage

• User assigns colors and opacities to a set of
  materials
• Map data values which been classified to material
  percentages
• Decide the color with material percentages

15
User interfaceStep sequences (1)
16
User interfaceStep sequences (2)
17
Outline

• Introduction
• Filter bank
• User interface
• Result Discussion
• Conclusion

18
Result (1)
19
Result (2)
20
Discussion

• Gaussian filters vs. Gradient magnitude
• Gaussian filters
• Indication of both the local and global structure
  at varying scales
• Tolerant of noise
• Gradient magnitude
• Regions with the most changes material
  boundaries
• Relate structures across scales

21
Outline

• Introduction
• Filter bank
• User interface
• Result Discussion
• Conclusion

22
Conclusion

• A new approach to interactive volume
  classification that allows the user to explore a
  filter bank space of inputs for specifying
  transfer functions
• Keeping the classification task simple, using
  small sets of 1D transfer function widgets
• Ability to classify volumetric features according
  to their sizes