Presenter: Cheong Hee Park Advisor: Victoria Interrante Texture Classification using Spectral...
-
date post
21-Dec-2015 -
Category
Documents
-
view
233 -
download
3
Transcript of Presenter: Cheong Hee Park Advisor: Victoria Interrante Texture Classification using Spectral...
Presenter: Cheong Hee Park
Advisor: Victoria Interrante
Texture Classification using Spectral Decomposition
OverviewGoal: Visualization of multivariate
data set in a planar 2D using principal perceptual features of texture.
Step1: Classify textures into meaningful categories.
• Classification by directionality• Classification by regularity• Structural grouping
Step2: Synthesize a series of textures to convey values of multivariate data.
Review of texture analysis and data visualization
Discrete Fourier TransformClassification by directionalityClassification by regularityClassification by StructureFuture work
Visualization of Magnetic field using orientation, size and contrast
Using Visual Texture for Information Display - Colin Ware and William Knight (1995)
Display over a 3D surface using height, density and regularity
Building Perceptual Textures to Visualize Multidimensional Datasets (C. Healey, J. Enns, 1998 )
Harnessing natural textures for multivariate visualization (Victoria Interrante)
farms(percent) in 1992 percent change of farms from 1987 to 1992
What is texture? An image composed of uniform or non-uniform
repetition of natural or artificial patterns
Methods used for texture analysis
• Autocorrelation• Co-occurrence based method• Parametric models of texture • Gray level run lengthSpectral decomposition
Principal features of texture
Directionality: directional vs
non-directional Coarseness: coarse vs fine Contrast: high contrast vs low contrast Regularity: regular vs irregular
(periodicity, randomness) Line likeness: line-like vs blob-like Roughness: rough vs smooth
Toward a texture naming system: identifying relevant dimensions of texture(A.R.Rao, G.L.Lohse, 1996)
Lace-like
Directional,Locally-oriented
Non-random,Repetitive,
non-directional <-> directional
Marble-like
Random,Non granular,
Somewhat repetitive
random
Random,granular
Texture features corresponding to visual perception -Tamura, Mori and Yamawaki
psychological measurement of directionality
(by human subjects using pair comparison method)
computational measurement of directionality
(using local vertical and horizontal directional operators)
Modeling spatial and temporal textures - Fang Liu
Decomposition of texture into three components based on Wold theory:
harmonic(periodicity),
evanescent(directionality),
indeterministic(random).
Measured deterministic energy from harmonic and evanescent components, and indeterministic energy from indeterministic component.
Instead of two processes FFT and local window interpolation, apply global sinusoidal filters directly to the texture
- Psychological experiment by Tamura- Ours(by interpolation)- (by direct filtering)- computational experiment by Tamura
Q: How can we judge which method is better ?
Future workHow to map attributes of multivariate data to texture perceptual dimensions independently? What perceptual features of texture are
most orthogonal?
-- Minimize interference when they are combined for display of multivariate data.
Mapping should be continuous within an attribute and make maximum distinction between attributes.