Combining Channels & 2D Scalar Fields

DAS732: Data Visualization

Jaya Sreevalsan Nair, IIIT Bangalore
August 22, 2024
Redundant
Encoding

2
http://www.nytimes.com/interactive/2013/05/25/sunday-review/corporate-taxes.html

3
https://www.nytimes.com/interactive/2014/10/19/upshot/peyton-manning-breaks-touchdown-passing-record.html

4
 5
Munzner, T. (2014). Visualization analysis and design. CRC press.
Combining
Channels

6
 Combining Channels

Can we combine multiple channels together in a single visual channel?

Munzner, T. (2014). Visualization analysis and design. CRC press. 7

https://hansjoerg.me/2019/02/15/plotting-many-groups-with-ggplot2/
 Combining Channels

Can we combine multiple channels together in a single visual channel?

Munzner, T. (2014). Visualization analysis and design. CRC press. 8

https://hansjoerg.me/2019/02/15/plotting-many-groups-with-ggplot2/
 Choice of Colors for Separability of Channels
ColorBrewer
Palettes
Palettes
Tableau

Wang, Yunhai, Xin Chen, Tong Ge, Chen Bao, Michael Sedlmair, Chi-Wing Fu, Oliver Deussen, and Baoquan Chen. "Optimizing color assignment for 9
perception of class separability in multiclass scatterplots." IEEE transactions on visualization and computer graphics 25, no. 1 (2018): 820-829.
 Choice of Foreground and Background Colors

Lu, Kecheng, Mi Feng, Xin Chen, Michael Sedlmair, Oliver Deussen, Dani Lischinski, Zhanglin Cheng, and Yunhai Wang. "Palettailor: Discriminable 10
colorization for categorical data." IEEE Transactions on Visualization and Computer Graphics 27, no. 2 (2020): 475-484.
 Scientific
Visualization:
Data
Representation

11
 Research Track Classification in Visualization

● Scientific Visualization – pertaining to scientific and engineering datasets

● Information Visualization – pertaining to datasets where spatial data is
computed as an abstraction
● Visual Analytics – pertaining to data science workflows involving
visualizations and feedback loop

Image courtesy : https://www2.cs.uh.edu/~chengu/Teaching/Fall2013/IntroVis_fall2013.html

12
Data Types

1. Continuous/Sampled Data: Analogous to temporal (time) dimension, sampling

is applied in physical spatial dimensions.
2. Multidimensional/Multivariate Data: Classification based on dimensionality.

Image courtesy:https://www.tutorialspoint.com/digital_communication/digital_communication_sampling.htm 13
1. Continuous and Discrete Data

Two types of discrete data: Structured vs Unstructured data ==

Gridded (rainbow colored surface) vs Scattered (red points).

14
Image courtesy:https://www.mathworks.com/help/matlab/ref/griddata.html
Gridding

Standard procedure done in computational methods to convert scattered points to

gridded data.

● Gridded data is easier to handle (pre-process/compute/visualize/etc.) using

indexing.

15
Sampling and Reconstruction

Nyquist-Shannon Sampling Theorem:

If a function x(t) contains no frequencies higher than B Hertz, it is completely

determined by giving its ordinates at a series of points spaced 1/(2B) seconds apart.
https://en.wikipedia.org/wiki/Nyquist%E2%80%93Shannon_sampling_theorem
16
 Sampling and Reconstruction

● Data acquisition provides samples.

● Resampling often occurs to get regular structured grids (or data).
● Understanding sampling is important to avoid aliasing (jagged artifacts) as well
as Moiré patterns.
● Nyquist-Shannon sampling theorem: states that a band-limited signal (i.e.,
restricted to a particular frequency range), that is sampled with a frequency at
least twice its highest frequency (Nyquist frequency),is completely determined
by its samples.

For further reference:

On sampling theory (in general):
http://graphics.cs.ucdavis.edu/~okreylos/TAship/Seminar/SamplingTheory101.pdf
Sampling for graphics: http://www.pbrt.org/chapters/pbrt_chapter7.pdf
Sampling for images:
http://graphics.cs.cmu.edu/courses/15-463/2011_fall/Lectures/SamplingReconstruction.pdf

Image Courtesy: By P. Fraundorf - Own work, CC BY-SA 3.0, https://commons.wikimedia.org/w/index.php?curid=6272747 17

 Sampling Artifacts - Moiré patterns

In mathematics, physics, and

art, moiré patterns or moiré
fringes are large-scale
interference patterns that can be
produced when a partially
opaque ruled pattern with
transparent gaps is overlaid on
another similar pattern.

Image Courtesy: http://www.cs.jhu.edu/~misha/MyPapers/3DV15.pdf ; GVCL

https://en.wikipedia.org/wiki/Moir%C3%A9_pattern
18
 Sampling Artifacts - but not Moiré patterns

Image Courtesy: https://blog.playment.io/list-of-lidar-datasets-for-autonomous-vehicles-till-2018/

19
2. Multidimensional Data

● Misconception between multidimensional and multivariate data.

● Right usage should be multidimensional multivariate data (mdmv data) where
(physical) dimensions are linearly independent variables and attributes are
dependent variables – this is an outdated terminology though.
○ Reference: Pak Chung Wong and R. Daniel Bergeron. 1994. 30 Years of Multidimensional Multivariate
Visualization. In Scientific Visualization, Overviews, Methodologies, and Techniques, Gregory M. Nielson,
Hans Hagen, and Heinrich Müller (Eds.). IEEE Computer Society, Washington, DC, USA, 3-33.

● Projecting from higher to lower dimensions may be achieved mathematically,

e.g. using subspace projection.
● However, extending concepts from lower to higher dimensions is not a trivial
or an intuitive task. Hence, this must not be applied mindlessly.
○ Four-Circle Paradox demonstrates the non-triviality.

20
Four Circle Paradox

In 2-d, in the square, the radius of

inner circle, r2 = √2 − 1,

Similarly in 3-d, in a cube, it is

r3 = √3 − 1,

Extending to n-dimensions in a
n-dimensional hypercube, the
radius is rn = √n − 1. Image Courtesy: http://bit-player.org/2011/the-n-ball-game

21
Four Circle Paradox

In 2-d, in the square, the radius of

inner circle, r2 = √2 − 1,

Similarly in 3-d, in a cube, it is

r3 = √3 − 1,

Extending to n-dimensions in a
n-dimensional hypercube, the
radius is rn = √n − 1.

Except that for n ≥ 10, the inner

radius rn > 2,
Image Courtesy: http://bit-player.org/2011/the-n-ball-game
which violates the boundary
condition of half length of
bounding hypercube in any
22
dimension being 2.
Fundamentals of Data Representation

● Continuous data comes from simulations; and is typically sampled to be

discretized.
● Sampled data is discrete, but falls under the category of “intrinsically”
continuous data.
● The large categories are: “intrinsically” continuous data and “intrinsically”
discrete data.
● There are no equivalents of each other is each other’s “domains”.

Continuous data may be modeled as a function f : D → C , where D ⊂ Rd and C

⊂ Rc ; f is called a d-dimensional (or d-variate), c-value function. f or its sampled
version is called a field.

23
Fundamentals of Data Representation

e.g. Consider temperature in a 3D volumetric ocean data, then d=3, c=1

(corresponding to temperature). If we considered both pressure and temperature
in the dataset, then c=2.

● Cauchy criterion: f is continuous if for every point p ∈ D, the following holds:

∀ ε > 0, ∃δ > 0, such that if ǁx − pǁ< δ, where x ∈ D, then ǁf(x) − f(p)ǁ < ε.

Triplet D = (D, C , f) defines a continuous dataset.

● Function values c are also called attributes.

● Dimensionality c of co-domain also called attribute dimension.
● 1 ≤ c ≤ 4 usually, for direct correspondence for graphical representation, e.g.
scalar (c=1), vector (c=2 or 3), color (c=3 or 4), tensor attributes.

24
 Fields

Scalar Vector Tensor

Zhou, L., Fan, M., Hansen, C., Johnson, C. R., & Weiskopf, D. (2022). A review of three-dimensional medical image visualization. Health Data Science, 2022.
25
Methods of Scalar Field Visualization

For 2-dimensional data:

● Color mapping
● Elevation mapping
● Contour mapping
http://homepage.ntu.edu.tw/~weitingc/fortran_lecture/Lecture_P_4_2Dplot.slides.html

For 3-dimensional data:

● Color mapping
● Direct volume rendering
● Isosurface extraction

26
2D Scalar Field
Visualization:
Color Mapping

27
2D Scalar Field Visualization

For 2-dimensional data:

● Color mapping
● Elevation mapping
● Contour mapping

http://homepage.ntu.edu.tw/~weitingc/fortran_lecture/Lecture_P_4_2Dplot.slides.html

28
 Grid
Generation

29
https://www.ncl.ucar.edu/Applications/rdm2grid.shtml
 Grid Generation

http://web.mit.edu/8.13/matlab/MatlabTraining_IAP_2012/AGV/DemoFiles/ScriptFiles/html/Part7_SlicesIsosurfaces.html

30
 Grid Generation
What are the values at the vertices of this 2D grid cell?
A B

D C

http://web.mit.edu/8.13/matlab/MatlabTraining_IAP_2012/AGV/DemoFiles/ScriptFiles/html/Part7_SlicesIsosurfaces.html

31
 Grid Generation
What are the values at the vertices of this 2D grid cell?
A (0.3) B (-0.1)

D (0.1) C (-0.3)
What is wrong with this image?
http://web.mit.edu/8.13/matlab/MatlabTraining_IAP_2012/AGV/DemoFiles/ScriptFiles/html/Part7_SlicesIsosurfaces.html

32
 Grid Generation
What are the values at the vertices of this 2D grid cell?
A B

D C

http://web.mit.edu/8.13/matlab/MatlabTraining_IAP_2012/AGV/DemoFiles/ScriptFiles/html/Part7_SlicesIsosurfaces.html

33
 Grid Generation
What are the values at the vertices of this 2D grid cell?
A B

O (0.0)

D C

http://web.mit.edu/8.13/matlab/MatlabTraining_IAP_2012/AGV/DemoFiles/ScriptFiles/html/Part7_SlicesIsosurfaces.html

34
 Color Mapping

Dataset: Simulation of
three-dimensional radiation
hydrodynamics of ionization front
instabilities.
2D slice through the data set of the
logarithm of density over time
demonstrating an instability.

Image courtesy: SciVis Contest Dataset Description

http://sciviscontest-staging.ieeevis.org/2008/data.html

Good sources of information for color mapping:

https://colorbrewer2.org/

https://matplotlib.org/3.1.0/tutorials/colors/colormaps.html

35
Color
Spectrum
Image Courtesy:
Matplotlib

36
Color
Spectrum
Image Courtesy:
Matplotlib

37
Color
Spectrum
Image Courtesy:
Matplotlib

38
Color Mapping Techniques

● Parametric mapping using linear interpolation

○ For a line segment PQ, any point in the interior of the line segment R= (1-
𝜶)P + 𝜶Q, where the parameter corresponding to R is

𝜶 = length(RP)/length(PQ).

○ Apart from position coordinates, all properties with linear behaviour can
use this formula: C(R)= (1-𝜶)C(P) + 𝜶C(Q).
● Color spectrum: 2-colour, 3-colour, n-colour (rainbow color spectrum)

Image courtesy: https://harmoniccode.blogspot.com/2011/04/bilinear-color-interpolation.html

39
 Color Mapping Techniques
● Parametric mapping using bilinear
interpolation.
A B
● Extending linear interpolation to 2D
space.
● For an interior point R in a rectangle
ABCD, walking clockwise, starting
with A being the left bottom corner,
parameters are
○ 𝜶 = distance(R,AB)/length(AB)
○ 𝜷 = distance(R,BC)/length(BC)

C(R) = (1-𝜶)(1-𝜷)C(A) + (1-𝜶)𝜷 C(B)+

𝜶𝜷 C(C) + 𝜶(1-𝜷 )C(D)
D C

40
Image courtesy: https://harmoniccode.blogspot.com/2011/04/bilinear-color-interpolation.html
2D Scalar Field
Visualization:
Elevation
Mapping

41
2D Scalar Field Visualization

For 2-dimensional data:

● Colour mapping
● Elevation mapping
● Contour mapping

http://homepage.ntu.edu.tw/~weitingc/fortran_lecture/Lecture_P_4_2Dplot.slides.html

42
 Elevation Mapping

Terrain of the Grand Canyon

https://www.cc.gatech.edu/projects/large_models/gcanyon.html

43
 Elevation Mapping

Terrain of the Grand Canyon

https://www.cc.gatech.edu/projects/large_models/gcanyon.html

2-dimensional view

3-dimensional view 44
How to Do Elevation Mapping?

Provide z-value (height) based on a function of (x,y). This gives a 2.5D surface.

45
Image courtesy: MATLAB
Challenges

● Mapping heights for a wide range of values

● Surface complexity is higher than curve complexity
● Intuition of elevation is not straightforward in all applications

46
 ● Redundant Encoding
● Combining Channels

Summary ● Introduction to Scientific

Visualization & Fields
● 2D Scalar Field Visualization
○ Color Mapping
○ Elevation Mapping

47