Module 6 – Simplified Notes

Module -6

Evaluating Visualization, Arrange Networks and Trees

Arrange tables: Categorical regions – Spatial axis orientation – Spatial layout

density, Arrange spatial data: Geometry – Scalar fields – Vector fields – Tensor fields.
Arrange

networks and t rees: Connections, Matrix views – Containment, Map color: Color theory,
Color

maps and other channels.

Arrange tables

The Big Picture

Figure 7.1 shows the four visual encoding design choices for how to arrange tabular data
spatially. One

is to express va lues. The other three are to separate, order, and align regions. The spatial
orientation

of axes can be rectilinear, parallel, or radial. Spatial layouts may be dense, and they may be
space -

filling

Why Arrange?

The arrange design choice covers all aspects of the use of spatial channels for visual
encoding.

It is the most crucial visual encoding choice because the use of space dominates the user’s

mental model of the dataset.

The three highest ranked effectiveness channels for quantitative and ordered attri butes
are all

related to spatial position: planar position against a common scale, planar position along an

unaligned scale, and length.

Arrange by Keys and Values

 The distinction between key and value attributes is very relevant to visually encoding tabl
e data.

A key is an independent attribute that can be used as a unique index to look up items in a
table,

while a value is a dependent attribute: the value of a cell in a table.

Key attributes can be categorical or ordinal, whereas values can be all three of the types:

categorical, ordinal, or quantitative.

The unique values for a categorical or ordered attribute are called levels, to avoid the
confusion

of overloading the term value.

Express: Quantitative Values

Using space to express quantitative attribut es is a straightforward use of the spatial

position

channel to visually encode data.

The attribute is mapped to spatial position along an axis.

Example scatterplots

The idiom of scatterplots encodes two quantitative value variables using both the vertical
and

horizontal spatial position channels, and the mark type is necessarily a point.

Scatterplots are effective for the abstract tasks of providing overviews and characterizing

distributions, and specifically for finding outliers and extreme values.

Scatter plots are also highly effective for the abstract task of judging the correlation
between two

attributes.

Figure 7.2. Scatterplot. Each point mark represents a country, with horizontal and
vertical spatial

position encoding the primary quantitative attrib utes of life expectancy and infant
mortality. The

color channel is used for the categorical country attribute and the size channel for
quantitative
population attribute. From [Robertson et al. 08, Figure 1c].

Figure 7.3. Scatterplots. (a) Original diam ond price/carat data. (b) Derived log-scale

attributes are highly positively correlated. From [Wickham 10, Figure 10].

Separate, Order, and Align: Categorical Regions

The use of space to encode categorical attributes is more complex than the simple case o f

quantitative attributes where the value can be expressed with spatial position.

Spatial position is an ordered magnitude visual channel, but categorical attributes have

unordered identity semantics.

The principle of expressiveness would be violated if th ey are encoded with spatial

position.

The semantics of categorical attributes does match up well with the idea of a spatial
region:

regions are contiguous bounded areas that are distinct from each other.

List Alignment: One Key

With a single key, separati ng into regions using that key yields one region per item.

The regions are frequently arranged in a one -dimensional list alignment, either
horizontal or

vertical.

EXAMPLE: BAR CHARTS

The well -known bar chart idiom is a simple initial example. Figure 7.4 sh ows a bar chart of

approximate weights on the vertical axis for each of three animal species on the horizontal
axis.

Stacked bar chart

A stacked bar chart uses a more complex glyph for each bar, where multiple sub -bars are

stacked vertically.

The length of the composite glyph still encodes a value, as in a standard bar chart, but
each

subcomponent also encodes a length -encoded value.

 Stacked bar charts show information about multidimensional tables, specifically a two -

dimensional table with two keys.

The composite glyphs are arranged as a list according to a primary key.

The other secondary key is used in constructing the vertical structure of the glyph itself.

Stacked bar charts are an example of a list alignment used with more than one key
attribute.

They support the task of lookup according to either of the two keys.

Example: Stream graphs

The stream graph idiom shows derived geometry that emphasizes the continuity of the

horizontal layers that represent the artists, rather than showing individual vert ical glyphs
that

would emphasize listening behavior at a specific point in time.

Example: Dot and Line Charts

The dot chart idiom is a visual encoding of one quantitative attribute using spatial
position

against one categorical attribute using point mark s, rather than the line marks of a bar
chart.

Matrix Alignment: Two Keys

Datasets with two keys are often arranged in a two-dimensional matrix alignment

the other along the columns, so a rectangular cell in the matrix is the region for the

item values one key is distributed alon g the rows and the other along

the rectangular cell in the matrix the reg ion for showing the item values.

Example: Cluster Heat maps

The idiom of heat maps is one of the simplest uses of the matrix alignment: each c ell is fully

occupied by an area mark encoding a single quantitative value attribute with color. Heat
maps

are often used with bioinformatics datasets.

Example: Scatterplot Matrix

A scatterplot matrix (SPLOM) is a matrix where each cell contains an scatter plot chart. A

SPLOM shows al possible pairwise combinations of attributes, with the original attributes as

the rows and columns. . In contrast to the simple hea tmap matrix where each cell shows
one

attribute value, a SPLOM is an example of a more complex matrix where each cell shows a

complete chart

Volumetric Grid: Three Keys

Just as data can be aligned in a 1D list or a 2D matrix, it is possible to align data in

three dimensions, in a 3D volumetric grid.

Recursive Subdivision: Multiple Keys

With multiple keys, it’s possible to extend the above approaches by recursively
subdividing the

cell within a list or matrix. That is, ordering and alignment is still used in the same way,
and

containment is added to the mix.

Spatial Axis Orientation

An additional design choice with the use of space is how to orient the spatial axes:
whether to

use rectilinear, parallel, or radial layout.

Rectilinear Layouts

In a rectilinear layout, regions or items are distributed along two perpendicular axes,
horizontal

and vertical s patial position, that range from minimum value on one side of the axis to a

maximum value on the other side. Rectilinear layouts are heavily used in vis design and
occur

in many common statistical charts. All of the examples above use rectilinear layouts.

Parallel Layouts

The rectilinear approach of a scatterplot, where items are plotted as dots with respect to
perpendicular axes, is only usable for two data attributes when high -precision planar
spatial

position is used. Even if the low-precision visual channel of a third spatial dimension is
used,

then only three data attributes can be shown using

spatial position channels.

Example: Parallel Coordinates

The idiom of parallel coordinates is an approach for visualizing many quantitative

attributes at

once usi ng spatial position. As the name suggests,the axes are placed parallel to each
other,

rather than perpendicularly at right angles.

Radial Layouts

In a radial spatial layout, items are distributed around a circle using the angle channel in

addition to one or more linear spatial channels, in contrast to the rectilinear layouts that
use

only two spatial channels.

Example: Radial Bar Charts

The same five -attribute dataset is encoded with a rectilinear bar chart in Figure 7.16(a) and

with a radial alternative in Figure 7.16(b). In both cases, line marks are used to encode a

quantitative attribute with the length channel, and the only difference is the radial versus
the

rectilinear orientation of the axes.

Example: Pie Charts

Pie charts encode a s ingle attribute with area marks and the angle channel. Despite their

popularity, pie charts are clearly problematic when considered according to the visual
channel

properties

Spatial Layout Density

Another design choice with spatial visual encoding idiom s is whether a layout is dense
or

sparse. A related, but not identical, choice is whether a layout is space -filling.

Dense

A dense layout uses small and densely packed marks to provide an overview of as many
items

as possible with very high information . A maximally dense layout has point marks that are
only

pixel-oriented . a single pixel in size and line marks that are only a single pixel in width.
The

small size of the marks implies that only the planar position and color channels can be
used in

visual encoding; size and shape are not available, nor are others like tilt, curvature, or
shape

that require more room than is available.

Example: Dense Software Overview

Space -Filling

A space -filling layout has the property that it fills all available space

in the view, as the name implies. Any of the three geometric possibilities discussed
above can

be space -filling. Space -filling layouts typically use area marks for items or containment
marks

for relationships, rather than line or connection marks, or point mark s.

Arrange Spatial Data

For datasets with spatial semantics, the usual choice for arrange is to use the given spatial

information to guide the layout. In this case, the choices of express , separate , order , and
align

do not apply because the position channe l is not available for directly encoding attributes.
The

two main spatial data types are geometry, where shape information is directly conveyed by
spatial elements that do not necessarily have associated attributes, and spatial fields,
where

attributes are associated with each cell in the field.

Why Use Given?

The common case with spatial data is that the given spatial position is the attribute of
primary

importance because the central tasks revolve around understanding spatial relationships.
In

these cases , the right visual encoding choice is to use the provided spa-tial position as
the

substrate for the visual layout, rather than to visually encode other attributes with marks
using

the spatial position channel.

This choice may seem obvious from common sens e alone. It also follows from the

effectiveness principle, since the most effective channel of spatial position is used to
show the

most important aspect of the data, namely, the spatial relationships between elements in
the

dataset.

Geometry

Geometric data does not necessarily have attributes associated with it: it conveys shape

information directly through the spatial position of its elements.

The field of computer graphics addresses the problem of simply drawing geometric data.

What makes geometry intere sting in a vis context is when it is derived from raw source

data as the result of a design decision at the abstraction level.

A common source of derived geometry data is geographic information about the Earth.

Geometry is also frequently derived from comp utations on spatial fields.

Geographic Data
Cartographers have grappled with design choices for the visual representation of
geographic

spatial data for many hundreds of years. The term cartographic generalization is closely
related

to the term Abstraction it refers to the set of choices about how to derive an appropriate

geometry dataset from raw data so that it is suitable for the intended task of the map users.

Example: Choropleth Maps

A choropleth map shows a quantitative attribute encoded as color over regions delimited
as

area marks, where the shape of each region is determined by using given geometry. The
region

shapes might either be

provided directly as the base dataset or derived from base data based on cartographic

generalization choices. The majo r design choices for choropleths are how to construct the

colormap, and what region boundaries to

use.

Figure 8.2 shows an example of US unemployment rates from 2008 with a segmented

sequential colormap. The white -to-blue colormap has a sequence of nine levels with

monotonically decreasing luminance. The region granularity is counties within states.

Scalar Fields: One Value

A scalar spatial field has a single value associated with each spatially defined cell.

Scalar fields are often collected through medi cal imaging, where the measured value is

radio -opacity in the case of computed tomography (CT) scans and proton density in the

case of magnetic resonance imaging (MRI) scans.

There are three major families of idioms for visually encoding scalar fields: sli cing, as

shown in Figure 8.3(a); is ocontours, as in shown Figure 8.3(b); and direct volume

rendering, as shown in Figure 8.3(c). With the is ocontours idiom, the derived data of

lower -dimensional surface geometry is computed and then is hown using standar d
computer graphics techniques: typically 2D isosurfaces for a 3D field, or 1D is olines for

a 2D field .

With the direct volume rendering idiom, the computation to generate an imag

from a particular 3D viewpoint makes use of all of the informatio n in the full 3D spatial
field.

With the slicing idiom, information about only two dimensions at once is shown as an
image;

the slice might be aligned with the original axes of the spatial field or could have an
arbitrary

orientation in 3D space.

Isocontou rs

A set of isolines , namely, lines that represent the contours of a particular level of the

scalar value, can be derived from a scalar spatial field.

Example: Topographic Terrain Maps

Topographic terrain maps are a familiar example of isolines in widespre ad use by the

general public. They show the contours of equal elevation above sea level layered on

top of the spatial substrate of a geographic map.

Example: Flexible Isosurfaces

The flexible isosurfaces idiom uses one more level of derived data, the simpl ified contour
tree,

to help users find structure that would be hidden with the standard single -level approach.
There

may be multiple disconnected isosurfaces for a given value: as the value changes, individual

components could appear, join or split, or dis appear. The contour tree tracks this evolution

explicitly, showing how the connected isosurface components change their nesting
structure.

Figure 8.5. A full contour tree with over 1.5 million edges does

not help the user explore isosurfaces. From [Carr et al. 04,
Figure 1].

Direct Volume Rendering

The direct volume rendering idiom creates an image directly from the information
contained

within the scalar spatial field, without deriving an intermediate geometric representation of
a

surface.

The algorithmi c issues involved in the computation are complex; a great deal of work
has been

devoted to the question of how to carry it out efficiently and correctly

Example: Multidimensional Transfer Functions

Vector Fields: Multiple Values

Vector field datasets are often associated with the application domain of computational
fluid

dynamics (CFD), as the outcome of

flow simulations or measurements. Flow vis in particular deals with a specific kind of
vector

field, a velocity field, that contains information about both direction and magnitude at each
cell.

The three common cases are purely 2D spatial fields, purely 3D spatial

fields, and the intermediate case of flow on a 2D surface embedded within 3D space. Time
-

varying flow datasets are called unsteady , as opposed to steady flows where the behavior

does not change over time.

One of the features of interest in flows are the critical points , the points in a flow field
where

the velocity vanishes. They are classified by the behavior of the flow in their
neighborhoods:

the three main types are attracting sources , repelling sinks , and saddle points that
attract from
one direction and repel from another. 2 Also, sources and sinks may or may not have
circulation

around them._ Figure 8.7 shows these five types of critical points .

Flow Glyphs

The flow glyph idioms show local information about a cell in the field using an object with

internal substructure; one of the most basic choices is an arrow .

Geometric Flow

The geometric flow idioms compute derived geometric data from the ori ginal field using

trajectories computed from a sparse set of seed points and then directly show the derived

geometry. One major algorithmic issue is how to compute the trajectories

Texture Flow

The texture flow idioms also rely on particle tracing, but with dense coverage across the
entire

field rather than from a carefully selected set of seed points._ They are most commonly
used

for 2D fields or fields on 2D surfaces.

Feature Flow

The feature flow vis idioms rely on global computations across the entire v ector field to
explicitly

locate all instances of specific structures of interest, such as critical points, vortices, and
shock

waves.

Tensor Fields: Many Values

Flow vis is concerned with both vector and tensor data. Tensor fields typically contain a
matrix

at each cell in the field, capturing more complex structure than what can be expressed in a

vector field. 3 Tensor fields can measure properties such as stress, conductivity, curvature,
and
diffusivity. One example of a tensor field is diffusion tensor d ata, where the extent to which
the

rate of water diffusion varies as a function of direction is measured with magnetic

resonance

imaging.

Arrange networks and trees

Figure 9.1. The node –link diagram family of visual encoding idioms uses the connection

chan nel, where marks represent links rather than nodes. The second major family of

network encoding idioms are matrix views that directly show adjacency relationships.
Tree structure

can be shown with the containment channel, where enclosing link marks show hi erarchical

relationships through nesting.

Connection: Link Marks

The most common visual encoding idiom for tree and network data is with node–

link diagrams , where nodes are drawn as point marks and the links connecting

them are drawn as line marks. This idiom uses connection marks to indicate the

relationships between

items.

Networks are also very commonly represented as node–link diagrams, using connection.
Nodes

that are directly connected by a single link are perceived as having the tightest groupin g,
while

nodes with a long path of multiple hops between them are less closely grouped. The
number of

hops within a path—the number of individual links that must be traversed to get from one
node

to another—is a network -oriented way to measure distances. W hereas distance in the 2D

plane is

a continuous quantity, the network -oriented distance measure of hops is a discrete

quantity. The
connection marks support path tracing via these discrete hops.

Node–link diagrams in general are well suited for tasks that involve understanding the
network topology :

the direct and indirect connections between nodes in terms of the number of hops
between them through

the set of links. Examples of topology tasks include finding all possible paths from one
node to another,

finding the shortest path between two nodes finding all the adjacent nodes one hop away
from a target

node, and finding nodes that act as a bridge between two components of the network that
would

otherwise be disconnected.

Node–link diagrams are most often laid out within a two-dimensional planar region.
While it is

algorithmically straightforward to design 3D layout algorithms.

One of the most widely used idioms for node–link network layout using connection
marks is

force -directed placement .

Matrix Views

Netwo rk data can also be encoded with a matrix view by deriving a table from the
original

network data.

EXAMPLE

ADJACENCY MATRIX VIEW

A network can be visually encoded as an adjacency matrix view, where all of the nodes
in

the network are laid out along the ver tical and horizontal edges of a square region and
links

between two nodes are indicated by coloring an area mark in the cell in the matrix that is

the intersection between their row and column. That is, the network is transformed into
the
derived dataset o f a table with two key attributes that are separate full lists of every node

in the network, and one value attribute for each cell records whether a link exists
between

the nodes that index the cell.

Figure 9.6(a) shows corresponding node–link and adjacenc y matrix views of a small

network. Figures 9.6(b) and 9.6(c) show the same comparison for a larger network .

Containment: Hierarchy Marks

Containment marks are very effective at showing complete information about hierarchical

structure, in contrast to co nnection marks that only show pairwise relationships between

two items at once.

Example: Tree maps

The idiom of tree maps is an alternative to node –link tree drawings, where the hierarchical

relationships are shown with containment rather than connection . All of the children of a

tree node are enclosed within the area allocated that node, creating a nested layout. The

size of the nodes is mapped to some attribute of the node. Figure 9.8 is a tree map view of

the same dataset as Figure 9.3, a 5161 -node com puter file system. Here, node size encodes

file size

Map Color and Other Channels

The colloquial term color is best understood in terms of three separate channels:
luminance, hue, and

saturation. The major design choice for colormap construction is wheth er the intent is to
distinguish

between categorical attributes or to encode ordered attributes. Sequential ordered

colormaps show a

progression of an attribute from a minimum to a maximum value, while diverging

ordered colormaps

have a visual indication of a zero point in the center where the attribute values diverge
to negative on
one

side and positive on the other. Bivariate colormaps are designed to

show two attributes simultaneously using carefully designed combinations of luminance,

hue, and

saturation .

Color Theory

Color is a rich and complex topic, and here I only touch on the most crucial aspects that
apply to vis.

Color Vision

The retina of the eye has two different kinds of receptors. The rods actively ontribute to
vision only in

low-light settings and provide low -resolution black and white information. I will thus not
discuss them

further in this book. The main sensors in normal lighting conditions are the cones . There
are three types

of cones, each with peak sensitivities at a different wavelength within the spectrum of

visible light. The visual system immediately processes these signals into three opponent
color channels :

one from red to green, one from blue to yellow, and one from black and white encoding

luminance information. The luminance chan nel conveys highresolution edge information,
while the red –

green and blue –yellow channels are lower resolution. This split between luminance and

chromaticity —what most people informally call would normally call color—is a central
issue in visual

encoding d esign.

The theory of opponent color channels explains what is colloquially called color
blindness , which affects

around 8% of men in its most common form. A more accurate term is color deficiency, since
a

“colorblind ” person ’s ability to differentiate colo r along the red–green channel is reduced
or absent but
their blue –yellow channel is still in full working order.

Color Spaces

The color space of what colors the human visual system can detect is three dimensional;
that is, it can

be adequately described using three separate axes. There are many ways to
mathematically describe

color as a space and to transform colors from one such space into another. Some of these
are extremely

convenient for computer manipulation, while others are a better match with the ch
aracteristics of human

vision.

The most common color space in computer graphics is the system where colors are
specified as triples

of red, green, and blue values, which is called the RGB system. Although this system is

computationally convenient, it is a very poor match for the mechanics of how we see. The
red, green,

and blue axes of the RGB color space are not useful as separable channels; they give rise to

the integral perception of a color. I

.Another color space, the hue–saturation –lightness or HSL system, is more intuitive and
is heavily used

by artists and designers.

The hue axis captures what we normally think of as pure colors that are not mixed with
white or black:

red, blue, green, yellow, purple, and so on. The saturation axis is the amount of white

mixed with that pure color. For instance, pink is a partially desaturated red. The lightness
axis is the

amount of black mixed with a color. A common design for color pickers is a disk with
white at

the center and the hue axis wrapped around the outside, with separate linear control for
the amount of
darkness versus lightness, as shown in Figure 10.2. The HSV space is very similar, where
V stands for

grayscale value and is linearly related to L.

Luminance, Saturation, and Hue

Luminance and saturati on are magnitude channels, while hue is a identity channel. When
I use the

generic term color , The magnitude channel of luminance is suitable for ordered data types.
However,

one consideration with luminance is our low accuracy in perceiving whether nonco
ntiguous regions

have the same luminance because of contrast effects.

The magnitude channel of saturation is also suitable for ordered data. Saturation shares
the problem of

low accuracy for noncontiguous regions. The number of discriminable steps for satu ration

is low: around three bins

Transparency

A fourth channel strongly related to the other three color channels is transparency :
information can be

encoded by decreasing the opacity of a mark from fully opaque to completely see-
through. Transparency

cannot be used independently of the other color channels because of its strong interaction
effects with

them: fully transparent marks cannot convey any information at all with the other three
channels.

Colormaps

A colormap specifies a mapping between colors and data values; that is, a visual encoding
with color.

Using color to encode data is a powerful and flexible design choice, but colormap design
has

many pitfalls for the unwary.

Figure 10.6 shows the taxonomy of colormaps; it is no coincidence that it mirrors the
taxonomy of data

types. Colormaps can be categorical or ordered , and ordered colormaps can be either
sequential

or diverging . Of course, it is important to match colormap to data type characteristics,

following the

expressiveness principle. Col ormaps for ordered data should use the magnitude
channels of luminance

and saturation, since the identity channel of hue does not have an implicit ordering.

Colormaps can either be a continuous range of values, or segmented

into discrete bins of color. _ Continuous colormaps are heavily used for showing
quantitative attributes,

especially those associated with inherently spatial fields. Segmented colormaps are

suitable for categorical data. For ordinal data, segmented colormaps would emphasize its
discrete n ature,

while continuous would emphasize its ordered nature. Bivariate colormaps encode two
attributes

simultaneously. While bivariate colormaps are straightforward to understand when the

second attribute

is binary, with only two levels, they are more difficult for people to interpret when both
attributes have

multiple levels.

Categorical Colormaps

A categorical colormap uses color to encode categories and group -

ings. Categorical colormaps are normally segmented. _ They a very effective when used
appropr iately;

for categorical data, they are the next best channel after spatial position.

Categorical colormaps are typically designed by using color as an integral identity

channel to encode a

single attribute, rather

than to encode three completely separate attributes with the three channels of hue,
saturation, and

luminance.

Ordered Color maps

An ordered colormap is appropriate for encoding ordinal or quantitative attributes. The

two major

variants of continuous colormaps for ordered data have expressiveness charactistics that
should match

up with the attribute type. A sequential colormap ranges from a minimum value to a
maximum value. If

only the luminance channel is used, the result is a grayscale ramp. When incorporating
hue, one end of

the ramp is a speci fic hue at full saturation and brightness. If saturation is the variable,
the other end is

pale or white; when luminance is the varying quantity, the other end is dark or black. A
diverging

colormap has two hues at the endpoints and a neutral color as a mi dpoint, such as white,
gray, or black,

or a high -luminance color such as yellow.

Bivariate Colormaps

The safest use of the color channel is to visually encode a single attribute; these
colormaps are known

as univariate . includes several colormaps that e ncode two separate attributes, called
bivariate . When one

of the two attributes is binary , meaning it has only two levels, then it is straightforward
to create a

comprehensible bivariate colormap with two families of colors by fixing a base set of hues
and varying

their saturation, as in the bivariate categorical –binary and diverging –binary examples

This approach can also be useful for a single categorical attribute that has a hierarchical
structure.
Colorblind -Safe Colormap Design

Designers using color sho uld take the common problem of red – green color blindness into
account. It is

a sex -linked inherited trait that affects 8% of males and a much smaller proportion of
females, 0.5%. In

the common forms of color blindness the ability to sense along the red –green opponent
color axis is

limited or absent. The problem is not limited to simply telling red apart from green; many
pairs that are

discriminable to people with normal color vision are confused, including red from black,
blue from

purple, light green from white, and brown from green.

Other Channels

Other identity channels are shape and motion. Textures and stippling use combinations
of multiple

channels.

Size Channels

Size is a magnitude channel suitable for ordered data. It interacts with most other chann
els: when marks

are too small, encodings in another channel such as shape or orientation simply cannot
be seen. Size

interacts particularly strongly with color hue and color saturation.

Length is one -dimensional (1D) size; more specifically, height is vert ical size and width is
horizontal

size. Area is two-dimensional (2D) size, and volume is three -dimensional (3D) . The
accuracy of our

perceptual judgements across these three channels varies considerably.

Angle Channel

The angle channel encodes magnitude information based on the orientation of a mark: the
direction that
it points. There are two slightly different ways to consider orientation that are essentially
the same

channel. With angle , the orientation of one line is judged with respect to another li ne.
With tilt, an

orientation is judged against the global frame of the display. _ While this channel is
orientation are often

used

somewhat less accurate than length and position, it is more accurate than area, the next
channel down

on the effectiveness ra nking.

Curvature Channel

The curvature channel is not very accurate, and it can only be used with line marks. It
cannot be used

with point marks that have no length, or area marks because their shape is fully
constrained. The number

of distinguishable bi ns for this channel is low, probably around two or three; it is in an

equivalence class

with volume (3D size) at the bottom of the magnitude channel ranking.

Shape Channel

The term shape is a catch -all word for a complex perceptual phenomenon.

Vision scien tists have identified many lower -level features that we can pre attentively
identify, including

closure, curvature, termination, intersection, and others. For the purposes of analyzing
visual encoding

with marks and channels, I simplify by considering shape as a identity channel that can be
used with

point and line marks. Applying shape to point marks is the common case, and is easy to
understand.

Applying the shape channel to line marks results in stipple patterns such as dotted and
dashed lines, as
discuss ed below. The shape channel cannot be applied to area marks because their shape
is constrained

by definition.

Motion Channels

Several kinds of motion are also visual channels, including direction of motion, velocity of
motion, and

flicker frequency . In order to use motion for visual encoding given a fixed spatial layout,
the typical

motion is a cyclic pattern where items somehow oscillate around their current location, so
that they do

not move outside the viewpoint, as would occur if they just continued to move in a

single direction.

Texture and Stippling

The term texture refers to very small -scale patterns. Texture is also a complex perceptual
phenomenon

that can be simplified by considering it as the combination of three perceptual dimensions:

Orientation , scale, and contrast. The first two pertain to the individual

Texture elements and have obvious mappings to the angle and size channels, respectively.
Contrast

refers to luminance contrast , which is related to the density of the texture elements; it
maps to the

luminance channel.

Texture can be used to show categorical attributes, in which case the goal is to create
patterns that are

distinguishable from each other using the combination of all three channels. In this

case, with sufficient care it is possib le to create at least one or two dozen distinguishable
bins.