Perceptual Display Hierarchies for Visualization

(1)

ABSTRACT

SAWANT, AMIT PRAKASH. Perceptual Display Hierarchies for Visualization. (Under the direction of Dr. Christopher G. Healey)

The advent of computers with high processing power has led to the generation of large,

multidimensional collections of data with increasing sizeanddimensionality. This has led to a critical need for ways to manage, explore and analyze large, multidimensional information

spaces. Visualization lends itself well to the challenge of exploring and analyzing these datasets

by managing and presenting information in a visual form to facilitate rapid, effective, and

meaningful analysis of data by harnessing the strengths of the human visual system.

Most visualization techniques are based on the assumption that the display device has

suf-ficient resolution, and that our visual acuity is adequate for completing the analysis tasks. However, this may not be true, particularly for specialized display devices (e.g., PDAs or

large-format projection walls). Our goal is to: (1) determine the amount of inlarge-formation a particular

display environment can encode; (2) design visualizations that maximize the information they

represent relative to this upper-limit; and (3) dynamically update a visualization when the

display environment changes to continue to maintain high levels of information content. A

collection of controlled psychophysical experiments were designed, executed, and analyzed

to identify thresholds for display resolution and visual acuity for four visual features: hue,

luminance, size, and orientation.

In computer graphics, level-of-detail hierarchies are often used to reduce geometric

com-plexity in situations where full resolution models are unnecessary. Using results from our

experiments, we applied similar logic to a visualization environment. If certain properties of

a dataset cannot be seen because their current visual representation falls below a resolution or

acuity threshold, they need not be included in the visualization. We built level-of-detail

per-ceptual display hierarchies to automatically add or remove information from a visualization as

(2)

devices. Our display hierarchies are combined with existing rules on the use of perception in

visualization to improve our ability to construct visualizations that are perceptually optimal for

a particular dataset, analysis tasks, and viewing conditions. We conclude by visualizing

mul-tidimensional weather data using a prototype visualization system that applies our perceptual

(3)

Perceptual Display Hierarchies for Visualization

by

Amit Prakash Sawant

A dissertation submitted to the Graduate Faculty of North Carolina State University

in partial fulfillment of the requirements for the Degree of

Doctor of Philosophy

Computer Science

Raleigh, North Carolina

2007

APPROVED BY:

Dr. Christopher G. Healey Dr. Steffen Heber

(4)

BIOGRAPHY

Amit Prakash Sawant was born to Prakash Shridhar Sawant and Priti Prakash Sawant on

July 19, 1977 in Baroda, India. He received a Bachelor of Engineering degree in Computer

Technology from Veermata Jijabai Technological Institute (VJTI) formerly know as Victoria

Jubilee Technical Institute, Mumbai, India. He worked with Tata Consultancy Services (TCS),

Mumbai, India for one year as a Systems Engineer. Amit joined the computer science masters

program at North Carolina State University in the spring of 2001. He received his Master of

Science in computer science in the spring of 2003 and since the fall of 2003 has been enrolled

(5)

ACKNOWLEDGMENTS

I would like to thank my advisor, advisory committee members, family, friends and

col-leagues for their help, motivation, and constructive criticism. Without their help and support

this dissertation would not have been possible. North Carolina State University’s Knowledge

Discovery Laboratory (KDL) provided an ideal working environment. I will miss being a part

of KDL.

Dr. Christopher Healey, my advisor, mentor and committee chair, whose guidance, support,

time, and energy have been instrumental toward my success in this research. His determination,

dedication, high standards, and critical reviews made sure my work was of outstanding quality.

It has been a pleasure working for Dr. Christopher Healey over the past few years and I cannot

thank him enough for the support and opportunity. I will miss working for him, going out for

lunch, and talking about sports, movies and politics.

Many friends and colleagues have helped me in various ways throughout the entire duration

of this research. They are - Ann, Anna, Andy, Arnav, Brent, Dan, Joe, Laura, Lloyd, Marivic,

Pragati, Ravi, Reshma, Sarat, Thomas, Vivek, and others whom I may have mistakenly

for-gotten to mention. A special thanks goes to Anna for her time, motivation, and critiques in

proofreading my drafts and listening to my practice presentations. I thank Arnav for his ideas

and technical support. I am grateful to Brent for his support and critical comments. Laura,

many thanks for proofreading my drafts and listening to my practice presentations. Sarat, I

am grateful to you for your insightful knowledge, technical support and constant motivation.

I thank Joe and Thomas for listening to my practice presentations and helping me setup the

projector for my defense. I thank Andy for his help with developing the perceptual display

hierarchy system. I thank all my friends and colleagues in KDL for listening to my practice

presentations and being a part of my research experiments. To my good friends, Ravi and

(6)

thank my friends and colleagues at Qimonda (formerly known as Infineon Technologies) and Network Appliance, Inc for all their help.

I would like to thank Dr. Steffen Heber, Dr. George N. Rouskas, and Dr. R. Michael Young

for agreeing to be on my advisory committee. Their suggestions and valuable comments have

been helpful in guiding my Ph.D. research. A special thanks also goes to Dr. Robert St. Amant

for agreeing to substitute for Dr. R. Michael Young at the last minute for my Ph.D. defense.

The love, encouragement, and blessings of my family made this dissertation possible. I

am truly indebted to my Mom and Dad (Priti Sawant and Prakash Sawant) for the sacrifices they made for me to attain higher education in the United States of America. I sincerely thank

my brother and sister (Vinit Sawant and Vrushali Sawant) for their good wishes and moral

support. Last but not the least, I would also like to thank my fiancee Sonali Parikh for her constant support and encouragement. I cannot thank her enough for being a part of my life,

and making everything seem possible, worthwhile and complete. I affectionately dedicate this

dissertation to them. Thank you!

“Dharma Matibhyah Udghrutah”- Bheeshma, Mahabharata. Dharma is born from reason,

Dharma is born from intellect,

Dharma is what your heart, your mind feels is right, Dharma is what your soul accepts as true and sacred.

Amit Prakash Sawant

(7)

List of Figures

1.1 Size and Visualization . . . 4

2.1 Structure of the Eye . . . 9

2.2 Calculation of Visual Angle . . . 10

3.1 Visualization Pipeline . . . 18

3.2 Scientific Visualization Example . . . 20

3.3 Information Visualization Example . . . 21

3.4 Tree-Map Example . . . 24

3.5 Three Level Filesystem Hierarchy . . . 25

3.6 Fisheye Lens Example . . . 26

3.7 Hyperbolic Browser Example . . . 27

4.1 Visualization of Performance Data . . . 28

4.2 Different Color Scales . . . 31

4.3 Coherency Visualization Technique . . . 32

4.4 Radial sweep from weather radar sensor . . . 34

4.5 Variation of Texture Dimensions . . . 36

5.1 Different Hue values . . . 43

5.2 Different Luminance values . . . 44

5.3 Different Size values . . . 44

5.4 Different Orientation values . . . 45

5.5 Display Resolution - Experimental Overview . . . 46

5.6 Hue Trial Example 1 . . . 48

5.11 Luminance Trial Example 1 . . . 53

(11)

5.16 Size Trial Example 1 . . . 58

5.21 Orientation Trial Example 1 . . . 63

5.26 aandrtagainst Target Present/Absent Trials . . . 73

5.27 aandrtagainst Display Resolution of elements . . . 74

5.28 aandrtagainst Target Feature . . . 75

5.29 a and rt for Target Present/Absent Trials against Display Resolution of the elements . . . 76

5.30 aandrtfor Target Present/Absent Trials against Target Feature . . . 77

5.31 aandrtfor Different Features against Different Display Resolutions . . . 78

6.1 Visual Acuity - Experimental Overview . . . 81

6.22 aandrtagainst Target Present/Absent Trials . . . 107

6.23 aandrtagainst Visual Angle of elements . . . 108

(12)

6.25 aandrtfor Target Present/Absent Trials against Visual Angle of the elements . 110

6.26 aandrtfor Target Present/Absent Trials against Target Feature . . . 111

6.27 aandrtfor Different Features against Different Visual Angles . . . 112

7.1 Visual FeatureVf’s Distinguishability Graph . . . 113

7.2 Distinguishability Graphs for Different Visual Features . . . 115

7.3 Visualization of the weather data at different display resolutions and visual angles withtemperaturemapped to hue . . . 121

7.4 Visualization of the weather data at different display resolutions and visual angles withtemperaturemapped to luminance . . . 122

7.5 Visualization of the weather data at different display resolutions with temper-aturemapped to size . . . 124

7.6 Visualization of the weather data at different visual angles with temperature mapped to size . . . 125

7.7 Visualization of the weather data at different display resolutions with temper-aturemapped to orientation . . . 127

7.8 Visualization of the weather data at different visual angles with temperature mapped to orientation . . . 128

7.9 Visualization of the weather data at different display resolutions with temper-ature → hue, pressure → luminance, wind speed → size, cloud coverage → orientation . . . 130

7.10 Visualization of the weather data at different visual angles withtemperature→ hue,pressure→luminance,wind speed→size,cloud coverage→orientation 131 7.11 Visualization of the weather data at different display resolutions . . . 133

7.12 Visualization of the weather data at different visual angles . . . 134

7.13 Visualization of the weather data at different display resolutions and visual angles136 7.14 Visualization of the weather data at a a display resolution of1×1pixels and a visual angle of0.0477◦ . . . 138

7.15 Visualization of the weather data at a display resolution of4×4pixels and a visual angle of0.1908◦ . . . 139

(13)

List of Tables

2.1 Some Basic Visual Acuities . . . 11

2.2 Display Resolution of Current Display Devices . . . 14

4.1 Different Visual Features used in Visualization . . . 30

4.2 Summary of Visual Feature Properties . . . 39

(14)

Chapter 1 Introduction

In recent years, advances in modern technology have made it possible for researchers and

sci-entists to collect large amounts of data. Data are growing at an incredible rate in the form of

paper, film, and electronic media. Some of the data types that contribute to this information overload are textual data, databases, image data, sensor data, and video data. However, our

ability to effectively represent and efficiently analyze this data has struggled to keep up with the increased storage capacities. It is difficult for viewers to identify important characteris-tics such as relationships, patterns, dependencies, boundaries, outliers, clusters, and anomalies

embedded within the underlying data. For raw data to be useful, it must be organized and

presented in a format that facilitates analysis and interpretation.

One promising approach to analyzing, exploring, and comprehending this information is

using visualization, an area of computer graphics that deals with the management and

presen-tation of information in a visual form to facilitate rapid, effective, and meaningful analysis of

data. A visualization harnesses the ability of the human visual system to analyze and interpret

large amounts of visual information at great speeds. It can be used to represent scientific (e.g., land and satellite weather information, geographic information systems, and molecular

(15)

1.1 Perception in Visualization

The knowledge of perception can be used to generate visualizations that harness the strengths of

the low-level human visual system and display data in ways that accentuate interesting items to

the user. Applying perceptual guidelines to “take full advantage of the available bandwidth of

the human visual system” has been cited as an important area of current and future research in

visualization [53, 61]. Exactly how a particular visualization technique and the display device

used to present the visualization affect the available visual bandwidth, however, is currently

unknown. Visualizations utilize visual channels to convey information to the observer. The

visual channels include features such as size, shape, hue, luminance, and texture. Independent

of the visual channels used, visual bandwidth depends on the screen’s size, both in terms of the

number of pixels and its physical dimensions [47]. This research aims to identify which

prop-erties of both the visualization and the display environment must be considered to maximize

the amount of visual information a user can perceive. Our overall goal is to study how these

properties affect the ability of a visualization to convey information to a user. We hypothesize

the available “visual bandwidth” depends, at least in part, on the following criteria:

1. The physical characteristics of the display device (e.g., resolution in terms of the total

number of pixels and the physical size of the display).

2. The acuity of the human visual system (e.g., the viewer’s ability to distinguish different

visual features such as color, orientation and size based on its absolute capabilities, and

on the visual angle subtended by an object on the viewer’s eye).

3. The visualization technique, that is, the methods used to map a data element’s values to

a visual representation.

4. The properties of the data (e.g., its dimensionality and number of elements) and the

(16)

To date, most research has focused on the last two criteria. Much less work has been conducted

on understanding how display resolution and visual acuity influence the effectiveness of a visualization. We propose to study thefirst two issues in detail and then combine the results with the last two criteria. A better understanding of these properties will help us perform

important tasks during visualization such as:

1. Provide fundamental knowledge required to construct perceptually salient visualizations.

2. Verify whether the display device properties and the limitations of the human visual

system can meet the requirements of a given visualization technique.

3. Characterize to what extent a visualization technique saturates “visual bandwidth”.

Most current algorithms simply assume that sufficient physical display properties are available for generating effective visualizations. However, there are now many different types of

dis-plays available to users: some with limited resources (e.g., the small size and low pixel count

on PDAs and mobile phones) and some with expensive capabilities (e.g., multi-projector

pow-erwalls, responsive workbenches, and high-resolution monitors). These displays are already

having a significant impact on visualization techniques, creating a strong need to learn how to use them effectively.

The acuity of the human visual system also introduces constraints on a visualization.

Im-provements in physical display properties alone cannot necessarilyfix these limits. For exam-ple, an on-screen element must subtend a minimum visual angle on the viewer’s retina to be

distinguishable. Increasing a display device’s pixel resolution (i.e., increasing pixels-per-inch

and therefore possibly decreasing the size of the on-screen elements) beyond a certain limit

may produce diminishing results in terms of perceiving different visual properties of a data

(17)

1.2 Size and Visualization

(a)

(b) (c)

Figure 1.1: Examples of visualizing with different viewing parameters: (top) a close-up of Hudson Bay, each square represents weather conditions on a 1₂◦longitude by 1₂◦latitude grid,temperaturemapped to hue (blue for cold to red for hot),pressuremapped to luminance (brighter for higher),wind speedmapped to size (larger for stronger),cloud coveragemapped to orientation (more tilted for denser), andprecipitationmapped to regularity (more irregular for heavier); (left) North America with allfive attributes visualized; (right) North America with onlytemperatureandpressurevisualized

Consider a simple example of visualizing a large, multidimensional dataset on a typical

CRT monitor, and assume that the viewer has zoomed in on a small subset of the dataset. In

this scenario a large number of pixels are available to display each data element. Thus, a fully

detailed visualization containing as many data attributes as can be shown effectively would

(18)

only a few pixels of screen space will be allocated to each data element, and thus many of the

visual features used to represent different data attributes may not be easy to distinguish. This

overloading of multiple visual features mapped to a data element with a limited number of

pixels could be counterproductive, since it may interfere with our ability to accurately identify

any data values at this low per-element resolution.

Figure 1.1 is a visualization of a weather dataset made up of monthly environmental and

weather conditions provided by the Intergovernmental Panel on Climate Change. The top

im-age displays weather data and all associated attributes of a small geographical area of North

America (Hudson Bay area), while both bottom images visualize the weather data for the entire

continent. In Figure 1.1atemperature,pressure,wind speed,cloud coverage, andprecipitation

are visualized using hue, luminance, size, orientation, and regularity of placement,

respec-tively. The bottom left image displays all the data elements and their data attributes. As only

a few pixels are available for each data element, many of the visual feature values are difficult to identify in this image. Moreover, the presence of certain features (e.g., small sizes) interfere

with our ability to see other features (e.g., color). In the bottom-right image the same elements

are visualized, but the number of attributes are reduced to two: temperature(visualized with hue) andpressure(visualized with luminance). Since both hue and luminance are distinguish-able even at small display resolutions, the underlying data patterns are easier to identify. The

differences in these images lead us to believe that our ability to comprehend data needs to be

adaptive to the display resolutions and physical sizes at which the data is displayed. To be

ef-fective at accomplishing this goal, we need to address important questions such as: (1) Which

data attributes should we visualize, and which visual features should we use? and (2) When

should visual features and their associated data attributes be removed from a visualization, and

when should they be reintroduced?

One possible approach is a visualization system that can smoothly reduce or increase the

(19)

For example, as the viewer zooms out, the number of pixels and the physical size per

ele-ment decreases, so attributes may need to be smoothly removed. Likewise additional attributes

can be rendered as the viewer zooms in. The idea is to maximize the utilization of the

dis-play’s capabilities and the viewer’s visual system in an effective and efficient manner, thereby maintaining a balance in the display environment (i.e., either more small elements with fewer

attributes encoded or fewer large elements with more attributes encoded). Zooming is

analo-gous to moving the visualization to a different display device with a different number of pixels,

different physical dimensions, or different viewing distances. In summary, techniques that

vary a visualization based on a viewer’s changing perspective can address how a visualization

should be updated both on a single display device and across different types of displays.

1.3 Research Objectives

Our objective is to study the requirements for a perceptual visualization hierarchy that

consid-ers physical device properties and visual acuity to maximize the information made available to

a viewer. This hierarchy will depend on the number of pixels needed for a visual feature to

rep-resent information effectively (i.e., what display resolutions) and the physical size needed for

our visual system to accurately identify and interpret a visual feature (i.e., what visual acuity).

Understanding the limits on display resolution and visual acuity will allow us to better

vali-date a given visualization technique and characterize the extent to which a technique saturates

“visual bandwidth”. This objective can be achieved by satisfying the following goals:

1. Investigate how display resolution and visual acuity define limits on the distinguisha-bility of commonly-used visual features such as hue, luminance, size, and orientation.

The results of these investigations,finding the minimum number of pixels and visual an-gle required to represent each feature, will help us validate a data-feature mapping and

(20)

2. Develop perceptual display hierarchies based on results from our display resolution and

visual acuity experiments. We will use these display hierarchies to construct effective

vi-sualizations and dynamically vary the visualization as the display environment changes.

3. Validate our theoreticalfindings with practical applications.

1.4 Thesis Organization

The material in this dissertation is organized as follows: In Chapter 2, we review visual

acu-ity and display device properties. Chapter 3 provides a general background on visualization.

Chapter 4 discusses different visual features. In Chapters 5 and 6, we present our controlled

display resolution and visual acuity experiments. Chapter 7 describes level-of-detail

percep-tual display hierarchies and demonstrates the application of our visualization techniques to real

datasets. Finally, Chapter 8 concludes and summarizes our study with discussions about future

(21)

Chapter 2 Visual Acuity and Display Properties

Physical limitations of the human eye and properties of display devices need to be taken into

consideration in order to design effective and efficient visualizations. In this chapter we discuss visual acuity and display device properties in detail.

2.1 Human Vision

Figure 2.1 shows the internal structure of the human eye. The important features are: the retina,

the lens, the fovea, the iris, the cornea, and the eye muscle. The lens focuses a small, inverted

image of the world onto the retina. The iris acts as a variable aperture, assisting the eye to

adjust to different lighting conditions. The retina consists of two types of photosensitive cells:

rods and cones. Cones are primarily responsible for color perception and rods are responsible

for intensity. Rods are typically ten times more sensitive to light than cones. There is a small

region at the center of the visual axis as shown in Figure 2.1 known as the fovea that subtends

1 or 2 degrees of visual angle. The structure of the retina is roughly radially symmetric around

the fovea. The fovea contains only cones, and linearly, there are about 147,000 cones per

(22)

Figure 2.1: Internal structure of a human eye

The human eye contains separate systems for encoding spatial properties (e.g., size,

lo-cation, and orientation) and object properties (e.g., color, shape, and texture). These spatial

and object properties are important features that have been successfully used by researchers in

psychology for simple exploration and data analysis tasks such as target detection, boundary

detection and counting, and by researchers in visualization to represent high-dimensional data

collections [67]. However, the limited number of rods and cones (about 120 million rods and

6 million cones) within the eye means that it can only absorb a certain amount of visual

infor-mation over a given period of time. Thus, even though we can generate images with a high

number of pixels-per-inch, it does not necessarily translate to an improvement in our analysis

abilities once the pixel-per-inch ratio surpasses the threshold beyond which pixels blur together

with their neighbors. This limitation thus poses an interesting question for the area of

(23)

color, texture, and motion in order for them to be perceptually identifiable?”

Two aspects of the human vision system that are of particular interest to us are display

resolution and visual acuity. Display resolution defines the number of discrete picture ele-ments (pixels) needed to accurately encode different values of a visual feature. Visual acuity

forms a set of guidelines on the visual system’s need for a minimum visual angle to recognize

differences in a feature.

2.1.1 Visual Angle

Figure 2.2: Visual angle subtended by an object on a human eye

Visual angle is the angle subtended by an object on the eye of an observer. Visual angles

are generally defined in degrees, minutes, and seconds of arc (a minute is ₆₀1 degree and a second is ₆₀1 minute). For example, a 0.4-inch object viewed at 22-inches has a visual angle of

approximately 1 degree. In Figure 2.2, visual angle can be calculated as [67]:

θ

2 = arctan(

ab

d ) (2.1)

The visual angle depends on two factors: (1) it is proportional to the actual size of the

(24)

2.1.2 Visual Acuity

Visual acuity is a measurement of our ability to see detail. Acuity is important because it

defines absolute limits on the information densities that can be perceived. Visual acuity for a person with 20/20 vision is measured as the minimum angle of the viewingfield that must be filled with an image to distinguish one feature from the rest of the image (measured in “minutes”, 20₂₀ = 1 minute [13]). Some of the basic acuities are summarized in Table 2.1 [67].

Table 2.1: Some basic visual acuities

Type Description

Point acuity (1 minute of arc) The ability to resolve two distinct point targets. Grating acuity (1-2 minutes of arc) The ability to distinguish a pattern of bright and

dark bars from a uniform gray patch.

Letter acuity (5 minutes of arc) The ability to resolve a letter. The Snellen eye chart is a standard way of measuring this ability. 20/20 vision means that a 5-minute letter target can be seen 90% of the time.

Stereo acuity (10 seconds of arc) The ability to resolve objects in depth. The acuity is measured as the difference between two angles for a just-detectable depth difference.

Vernier acuity (10 seconds of arc) The ability to see if two line segments are collinear.

Most of the acuity measurements in Table 2.1 suggest that we can resolve visual

phenom-ena, such as the presence of two distinct lines, down to about 1 minute (₆₀1 ◦) of visual angle.

This is in rough agreement with the spacing of receptors in the center of the fovea. For us to

see that two lines are distinct, the blank space between them should lie on a receptor; therefore,

we should only be able to perceive lines separated by roughly twice the receptor spacing.

How-ever, there are a number of superacuities, like stereo acuity and vernier acuity, which allow us

to perceive visual properties of the world to a greater precision than could be achieved based

on a simple receptor model.

Postreceptor mechanisms are capable of integrating the input from many receptors to obtain

(25)

to judge the collinearity of twofine line segments. This can be done with amazing accuracy to better than 10 arc second. The resolution of the eye is often measured in cycles per degree and

ranges from 1₂ arc minute (120 cycles/degree) to 1 arc minute (60 cycles/degree). Resolution

of 1 arc minute allows one to distinguish detail of 0.01 seconds at 3 feet. Consider a display

screen that is 20-inches wide and positioned 22-inches from the viewer. How many pixels

across one scanline subtending45◦ would it take to match human visual acuity? If we assume

human visual acuity to be 1₂ arc minute, then we would need120×45= 5400 pixels to match

our visual ability [12].

Neural postprocessing can efficiently combine input from two eyes. The area of the overlap is approximately120◦ with 30-35◦ monocular vision on each side. Combined horizontal FOV

is 180-190◦and vertical FOV is 120-135◦ for both eyes [12]. This suggests that if the data

ele-ments in a visualization environment lie within this region of overlap they are identified more accurately than the data elements that lie in the monocular region. Campbell and Green [23]

found that binocular viewing improves acuity by 7% as compared with monocular viewing.

2.2 Display Device Properties

Properties of a display device can have a significant effect on the capabilities of a visualization. However, precisely characterizing this effect has so far been limited in its ability to address

issues such as: (1) What fraction of a dataset can a display device represent effectively? and

(2) What fraction of a display can a viewer attend to at any given time? A good visualization

technique should take into account the display resolution, physical size, and viewing distance

in order to maximize both the quantity and the quality of the information it displays. We will

(26)

2.2.1 Display Resolution

A display device’s resolution defines the number of pixels it contains, expressed in the hor-izontal and vertical directions. It is also referred to as the physical resolution of a display device. The sharpness of the display depends on its resolution and on its physical size. The

same resolution will be sharper on a smaller monitor compared to a larger monitor because

the same number of pixels are spread out over a larger physical region [11]. We use the term

display resolution to refer to the resolution of an element on a particular display device (i.e., the number of pixels allocated to an element on the screen). Real-world data are visualized

on a range of display devices such as computer monitors (traditional CRTs and LCDs), PDAs,

mobile phones, and powerwalls. Table 2.2 shows common display resolutions for these types

of display devices [5, 6, 7, 8, 9, 4, 1].

The low physical resolution of devices like mobile phones and PDAs limits the amount of

information they can display at any given time. A common physical resolution for a PDA is

240×320 pixels at 3.5-inches diagonal. Consider the example of visualizing a large dataset

on a PDA screen. This would allocate very few pixels to each data element. Even if the

resolution is increased dramatically (i.e., a significant increase in pixels-per-inch), it would not fully resolve the issue. An element must subtend a minimum visual angle on the viewer’s

retina to be distinguishable. Increasing pixels-per-inch beyond a certain point will produce

diminishing results in terms of the amount of additional information a viewer can see. At the

opposite extreme, a large display device such as a powerwall typically results in a largefield of view (FOV). But, there is a limitation on the amount of information human eyes can perceive

based on the horizontal and vertical FOV. Also, as the FOV increases users have to utilize their

peripheral vision [18], and it is a known fact that static visual features do not perform well

under peripheral conditions [16].

(27)

Table 2.2: Display Resolution of current display devices

Display Device Manufacturer Model Resolution Screen Size

Mobile Phone Vodofone Sharp GX30 858×1144 2.2-inch screen

Apple iPhone 480×320 3.5-inch screen

Nokia N81 8GB 240×320 2.4-inch screen

Motorola RAZR V3a 176×220 2.1-inch screen

Nokia 6200 128×128 27.3×27.3mm

PDA Toshiba e805(BT) 800×600 4.0-inch screen

Sony Clie PEG-UX50 480×320 4.0-inch screen

AT&T Tilt 240×320 2.8-inch screen

HP iPAQ 610 240×320 2.8-inch screen

T-Mobile Sidekick iD 240×160 2.75-inch screen

Monitors Auto Vision Inc AVHRPC703 640×480 7.0-inch screen

COMPAQ MV520 800×600 15.0-inch screen

ViewSonic VX510 1024×768 15.0-inch screen

Sony CPD-E240 1280×1024 17.0-inch screen

BenQ FP94VW 1440×900 19.0-inch screen

HP W2007v 1680×1050 20.0-inch screen

ViewSonic VX2435wm 1920×1200 24.0-inch screen

Dell UltraSharp 2707WFP 1920×1200 27.0-inch screen

PowerWall SGI Onyx2 6400×3072 8×2.85m

POWER ONYX 3200×2400 6×8feet

Exec CUG 2560×2048 14×10feet

6400×3072. Nowadays, tiled displays and visualization walls are common and their display

resolutions can vary to a large extent depending on the number of tiles and resolution of each

tile. For a particular display resolution, it is important to determine which visual features can be

rapidly identified and which cannot, based on the number of the pixels that need to be allocated to each visual feature to make its values distinguishable.

2.2.2 Physical Size

Physical size is an important cue to sensory and judgment processes in humans. In a series of

(28)

21-inch monitors as compared to smaller ones. Chapanis and Scarpa [26] conducted experiments

comparing the readability of physical dials at different distances to examine the psychophysical

effects of distance and size. They used dials of different sizes and markings that were

propor-tional to the viewing distance so as to keep visual angles constant. They found that beyond

28-inches, dials were read more easily. The effects they found, however, were relatively small.

Studies conducted by Desney et al. [32] suggest that users performed better on spatial

orientation tasks that require mental rotation on large displays compared to desktop monitors.

The visual angle was held constant by adjusting the viewing distance to each of the displays.

Large displays provide users with a greater sense of presence, allowing them to imagine

rotat-ing their bodies within the environment. Smaller displays force users to imagine rotatrotat-ing the

environment around themselves [25, 70]. Large displays normally cast a larger retinal image,

offering a wider FOV. Czerwinski et al. [28] reports that a wider FOV increases the sense of

presence and improves performance in 3D navigation tasks, many of which are important in

visualization.

The physical size of a display device has a direct effect on the available FOV. Also, for

a fixed pixels-per-inch, larger display devices have higher resolutions and therefore may be capable of visualizing more information.

2.2.3 Viewing Distance

The standard distance to the viewer from the computer screen is approximately 22-inches [67].

For large displays such as powerwalls, the optimal viewing distance is about twice the width of

the display [3]. As the viewing distance increases, the FOV decreases. For example, a 16-inch

display placed 22-inches from the user produces a FOV of approximately40◦. Increasing the

(29)

Chapter 3 Visualization

Various researchers have defined visualization as follows:

1. “Visualization is the use of computer-supported, interactive, visual representations of

data to amplify cognition” [24].

2. “Visualization is a powerful link between the two most powerful information processing

systems: the human mind and the modern computer” [33].

3. “Visualization simply means presenting information in pictorial form” [33].

4. “Visualization is the transformation of data to a format amenable to understanding by the

human perceptual system while maintaining the integrity of the information” [37].

5. “Visualization is the use of computer imaging technology as a tool for comprehending

data obtained by simulation or physical measurement” [37].

Visualizations allow users to see, explore, and understand large amounts of information at

once. Visual representations translate data into a visible form that highlights important features,

including commonalities and anomalies. These visual representations make it easy for users to

(30)

perceptual reasoning through visual representations permits the analytical reasoning process to

become faster and more focused [63].

3.1 Advantages of Visualization

The potential advantage of visualization techniques over other techniques from statistics,

ma-chine learning, and artificial intelligence is that in many instances, it allows direct interaction by the user and provides immediate feedback. It also supports user steering, which is difficult to achieve in other non-visual techniques [15]. Graphical presentations allowquery-by-attention, that is, they answer the user’s questions by controlling the user’s visual attention (and

review-ing immediate feedback) rather than forcreview-ing the user to manipulate a data query interface and

wait for a server to return results [34].

3.2 Visualization Process

Consider a datasetD ={e1, .., en}containingn sample points, or data elements,ei. A

multi-dimensional dataset represents two or more data attributes, A = {A1, ..., Am}, m > 1. The

data elements encode values for each attribute: ei ={ai,1, ..., ai,m},ai,j ∈Aj. A data-feature

mapping converts the raw data into images that can be presented to a viewer. Such a mapping

is denoted by M = (V,Φ), whereV = {V1, ..., Vm} is a set ofm visual featuresVj selected

to represent each attribute Aj, and Φj : Aj → Vj maps the domain of Aj to the range of

displayable values in Vj. Visualization is thus the selection of M and a viewer’s ability to

comprehend the images generated by M. An effective M must produce images that support

rapid, accurate, and effortless exploration and analysis [43]. The actual process for visualizing

(31)

Figure 3.1: Visualization Pipeline

3.2.1 Visualization Pipeline

The broad operations within a visualization pipeline, as shown using Figure 3.1 include:

1. Converting raw data to a data table, which can then be mapped to a visual structure.

2. Applying view transformations to increase the amount of information that can be

visual-ized.

3. Interacting with these visual structures and the parameters of the mappings to create an

information workspace for visual sense making [24].

With all types of visualization, the hope is that the human visual system can be used to rapidly

(32)

Eick developed many special-purpose visualization systems for non-geometric datasets. He

proposed the following guidelines to provide a foundation for designing visualizations [33].

• Ensure that the visualization is focused on the user’s needs by understanding the data

analysis task.

• Encode data using color and other visual characteristics.

• Facilitate interaction by providing a direct manipulation user interface.

• Show the evolution of temporally oriented data using animation.

3.3 Types of Visualization

Visualization techniques are loosely divided into two broad categories: scientific visualization and information visualization. Scientific visualization techniques deal with physically based data that possess some inherent geometry, while information visualization techniques deal with

abstract data. For example, a weather dataset with the attributes: latitude, longitude, tempera-ture, pressure, precipitation, windspeed, frost, cloud coverage, radiationandwet day frequency

is a scientific dataset. A movie database with the attributes: movie title, genre, length,yearof release anduser ratingis an information dataset. It is possible to remove the inherent geometry in a scientific dataset and use information visualization techniques to display its data elements. Similarly, we can add geometry to an abstract dataset and apply scientific visualization tech-niques to represent its contents. The use of an appropriate visualization system is determined

by the task and the format of the dataset. At its core, visualization is the conversion of numbers

(33)

Figure 3.2: Visualization of a weather dataset using perceptual texture elements withtemperature→hue,wind speed→density,pressure→size,precipitation→orientation, andcloud coverage→luminance

3.3.1 Scientific Visualization (SciVis)

Figure 3.2 is a visualization of a weather dataset, in which individual weather readings (or data

elements) are visualized using stroke glyphs (2D rectangular objects) that vary individual color

(34)

Figure 3.3: Visualization of a movie data using 3D glyphs

3.3.2 Information Visualization (InfoVis)

Information visualization, sometimes called InfoVis, is the interactive visual representation of

data with no inherent spatial mapping. Card et al. define information visualization as “the general application of assembling data objects into pictures, revealing hidden patterns” [24];

Herman et al. define it as “the visualization and navigation of abstract data structures” [45]. Information visualization also involves exploratory data analysis, a process of sifting through

data in search of interesting information or patterns [31]. The interactive nature of the

visual-ization allows for a constant redefinition of goals based on insights into the data [2].

Figure 3.3 shows a query result containing recommendations for movies [57]. We represent

the following attributes: movietitle,genre of the movie,year the movie was released,length

(35)

ranking values are positioned near the center of the spiral, and elements with low ranking

values are positioned near the periphery. Height represents predicteduser rating(short for low ratings to tall for high ratings). A gray-scaleflag at the bottom of the glyph representsyearof release ranging from 1921 to 2006 (dark for older movies to white for the most recent movies).

Light brown flags wrapped around the glyph at different heights represent genre (the order of the flags from bottom to top represent “Action”, “Comedy”, “Drama”, and “Romance”, respectively). Luminance represents the lengthof the movie (dark red for short to bright red for long). Predicteduser ratingwas also used as the ranking attribute to position glyphs along the spiral.

3.4 Visualization Techniques

There are a number of well-known techniques for visualizing non-spatial datasets. These

tech-niques can be roughly classified as geometric projection, iconic display, hierarchical, graph-based, pixel-oriented, and dynamic (and combinations thereof) [48]. In this section, we discuss

a few relevant visualization techniques.

In general, a technique is characterized by the data type being visualized, the display

al-gorithm being applied, and the interaction and distortion operations available to the viewer.

Datasets are normally composed of large numbers of elements. As the size grows,

visualiz-ing the dataset in its entirety becomes increasvisualiz-ingly difficult. A number of novel approaches have been proposed to address this problem, for example,focus + contextandoverview + de-tailalgorithms [30]. While both these approaches are hierarchical in nature, they differ in the methods they use to represent both the global structure and local details of a dataset.

Overview + detail techniques represent multiple views in separate windows. This main-tains the original spatial relationships without any distortion. Overview reduces search, allows

(36)

analysis [24]. The users may also need detailed information about local areas of the dataset.

Multiple views can be presented to the user in two ways: one at a time (time multiplexing) or

together in a single display in different regions of the screen (space multiplexing) [24].

Focus + context techniques provide users with a global overview as well as local detail information. Focus + contexttechniques present both types of information to the user based on the assumption that the user needs to view both global and local information simultaneously.

The content of information of each type needed by the user may not be the same however.

This leads to a selective reduction of information from certain parts of the display in order to

allocate more resources to correspond to the user’s interest.

Of the many visualization, interaction and distortion or reduction techniques, only a few

are explained here to give the reader some perspective.

3.4.1 Tree-maps

Tree-maps are a well-knownoverview + detail system. Tree-maps decompose a dataset into a rectangular image whose individual regions are hierarchically partitioned based on different

attributes within the dataset [59]. Tree-maps can display high levels of detail by zooming into

regions of interest. Later revisions to the tree-map allow users to select individual regions. This

expands the region tofill the screen and show a higher level of detail (fraction of the dataset), but at the expense of maintaining a view of the region’s location and context within the dataset.

Thus, treemaps either display low-levels or high-levels of detail, but not both at the same time.

Figure 3.4 shows the use of a tree-map to view the contents of afile system. The tree-map visualizes afilesystem with 15 files at three levels. The map is initially segmented into eight blocks (horizontal partitions). The top two blocks of thefirst level are divided again into four and three partitions respectively (vertical partitions), producing a new level. The dark red block

(37)

second level

third level

second level first level

1 2 3

4

5

6

7 8 9

10

12

13

15

11

14

(38)

Figure 3.5: A tree representation of the three levelfilesystem hierarchy shown in Figure 3.4

form a third level. Figure 3.5 shows the three level hierarchy.

3.4.2 Fisheye Lens

Thefisheye lens presents two levels of detail simultaneously. The entire dataset is displayed at a low level-of-detail. An interactive lens that zooms in about its center provides a high

level-of-detail display of the data directly beneath the lens [35]. The fisheye technique is a distorted-view approach that supports multiple focus points (via multiple fisheyes), enhances continuity through a smooth transition between overview and detail, and maintains location

constraints to reduce a user’s sense of spatial disorientation. This technique provides a balance

between local detail and global context. In a fisheye technique each element in a dataset is assigned a degree of interest (DOI), or priority, based on a priori interest and on distance from

the user’s current focus. The data is displayed only if the perceived value is greater than a user

defined threshold [17]. The use of multiple fisheyes distorts the image and thereby the user loses the overview context of the visualization.

Figure 3.6 shows the Washington, D.C. Metro system. The neighbors closest to the region

(39)

Figure 3.6: Fisheye lens view of the subway system in Washington, D.C.

the periphery are displayed with very little detail, but still provide a hint at the global structure.

3.4.3 Hyperbolic Browser

The hyperbolic browser structures information in a dataset as a tree embedded in the surface

of a sphere [50]. In a manner similar to a traditionalfisheye lens, a portion of the sphere facing outward uses hyperbolic geometry to form a lens, zooming the information being displayed as

the sphere is rotated. When a hyperbolic browser is used to visualize a tree, there are a large

number of nodes (the global overview) pushed near the circumference of the disc. Figure 3.7

shows that as a node moves to the center of the disc, more detailed information about the node

is visible due to the availability of more display space. Similar to thefisheye lens, a hyperbolic browser is a distorted-view approach, and therefore the viewer loses the low-level overview

context of the visualization.

(40)

low-(a) (b)

Figure 3.7: (a) a hyperbolic browser representing a family’s genealogy; (b) the old root node (in red) is moved out of focus and a new root brought into focus (in green)

(41)

Chapter 4 Visual Features

Figure 4.1: Visualization of storage controller performance data - each glyph represents one second of an NFS write workload to a storage controller;timestamp→linear radial position,disk data written→hue,disk busy→

luminance,nfs write latency→height,nfs write ops→size,disk data read→orientation

Figure 4.1 is an example of one glyph-based multidimensional visualization technique. It

represents the visualization of a workload of NFS writes to a storage controller. NetApp1

(42)

all response times. Dirty data is periodically asynchronous flushed to disk in an event called a consistency point (CP). A CP increases CPU and disk utilization. Every second of NFS

write workload (shown as a single data element) is visualized using a 2D glyph along a

space-filling spiral that varies its individual color and texture properties according to the data values it represents. Timestamp, disk data written, disk busy, nfs write latency, nfs write ops, and

disk data readare visualized using radial position, hue, luminance, height, size, and orienta-tion, respectively.

From Figure 4.1, the periodicity of CPs and their correlation with NFS throughput,

re-sponse time, and disk utilization is evident. A repeating pattern of several dark, low, large

samples (higher NFS throughput, lower NFS response time, lower disk utilization), followed

by a bright, tall, red sample (higher response time, higher disk read utilization), followed by

several large bright blue samples (higher disk write utilization) is visible. This suggests that

the beginning of a CP is accompanied by a small, temporary increase in NFS response time

and a burst of disk reads, probably for metadata needed to complete the CP. The majority of

the CP is spent doing disk writes as buffered data is committed to disk. Several unexpected

artifacts are also evident from the visualization. Most notably, several CPs include small, low

samples and tall, wide samples, indicating lower response time/lower throughput and higher

response time/higher throughput respectively. We expected an inverse relationship between

these attributes, so this may be indicative of variance in the workload being presented by the

client [58].

When we design a visualization, properties of the dataset and the visual features used to

represent its data elements must be carefully controlled to produce an effective result.

Impor-tant characteristics that must be considered include: (1) dimensionality (number of attributes

in the dataset), (2) number of elements, (3) visual-feature salience (strengths and limitations

that make it suitable for certain types of data attributes and analysis tasks), and (4) visual

(43)

this must be controlled or eliminated to guarantee effective exploration and analysis).

Using the knowledge of the above perceptual guidelines, we chose visual features that are

highly salient, both in isolation and in combination. We mapped features to individual data

attributes in ways that draw a viewer’s focus of attention to important areas in a visualization.

The ability to harness the low-level human visual system is attractive, since: (1) high-level

exploration and analysis tasks are rapid and accurate, (2) analysis is display size insensitive,

and (3) different features can interact with one another to mask information; psychophysical

experiments allow us to identify and avoid these visual interference patterns.

Effectively mapping data attributes to visual features requires better understanding the

properties of the visual features themselves. In the following sections, we will discuss the

properties of different visual features in more detail.

Table 4.1: Different visual features used in visualization

Feature Author

line (blob) orientation Jul´esz & Bergen (1983); Wolfe (1992)

length Triesman & Gormican (1988)

width Jul´esz (1984)

size Triesman & Gelade (1980)

curvature Triesman & Gormican (1988)

number Jul´esz (1985); Trick and Pylyshyn (1994)

terminators Jul´esz & Bergen (1983) intersection Jul´esz & Bergen (1983)

closure Enns (1986); Triesman & Souther (1986)

color (hue) Triesman & Gormican (1988); Nagy and Sanchez (1990); DZmura (1991)

intensity Beck et al. (1983); Triesman & Gormican (1988)

flicker Jul´esz (1971)

direction of motion Nakayama & Silverman (1986); Driver and McLeod (1992) binocular lustre Wolfe & Franzel (1988)

stereoscopic depth Nakayama & Silverman (1986)

3D depth cues Enns (1990)

(44)

4.1 Properties of Visual Features

A variety of visual features have been used in visualization. Some of them are listed in Table 4.1

[42]. In this chapter, we describe what has been learned about the visual features color, texture,

and motion in computer graphics and psychophysics. We discuss these properties based on:

1. domain: quantitative data (can perform arithmetics), ordinal data (obeys ordering rela-tions), or nominal data (equal or not equal to other values)

2. spatial frequency: low spatial frequency (i.e., data with low spatial variations in its val-ues) or high spatial frequency (i.e., data with sharp spatial variations in its valval-ues)

3. visual interference: whether one feature masks the presence of another; thus the most important data attributes should be displayed using the most salient visual features

4.2 Color

(45)

Color is a visual feature commonly used in visualization. An individual color can be

de-scribed by specifying a hue, saturation, and luminancevalue. Hue is the wavelength we see when viewing light of a given color. Saturation describes the strength of the color (i.e., its

distance from gray). Luminance refers to the intensity or the brightness of a given color. In this

thesis, we refer to a color by its hue name. Figure 4.2 shows examples of simple color scales,

such as the rainbow spectrum, red-blue or red-green ramps, and the gray-red saturation scale.

Figure 4.3: An example of Ware and Beatty’s coherency visualization technique, the four clouds of simi-larly-colored squares represent four coherent groups of data elements

4.2.1 Hue

One example use of color was proposed by Ware and Beatty to display correlation in a fi ve-dimensional dataset [27]. Each of the five data attributes is mapped to one of the following visual features: position along thex-axis, position along they-axis,redcolor,greencolor, and

(46)

Results show that the preattentive nature of a color depends on the saturation and size of

the color patch as well as the degree of difference from its surrounding colors. As a rule of

thumb, to avoid small-field color blindness, 1₂◦ of visual angle is probably the minimum size for color-coded objects [67]. One of the limitations of using color as a visual feature is that

humans are almost colorblind in their peripheral vision [71]. Healey showed that at most seven

hues can be rapidly distinguished from one another in a display [39]. Hue is best suited to

represent low spatial frequency nominal data.

4.2.2 Luminance

Luminance is a physical measure used to define the amount of light in the visible region of the electromagnetic spectrum [67]. Sophisticated techniques divide color along dimensions like

luminance, hue, and saturation to better control the difference viewers perceive between

dif-ferent colors. Researchers in visualization have combined perceptually balanced color models

with nonlinear mappings to emphasize changes across specific parts of an attribute’s domain, and have also proposed automatic colormap selection algorithms based on an attribute’s spatial

frequency, continuous or discrete nature, and the analysis tasks to be performed. Experiments

have shown that in order to select discrete collections of distinguishable colors we need to take

into consideration the following criteria [39, 41]:

1. color distance:the Euclidean distance between different colors as measured in a percep-tually balanced color model

2. linear separation: the ability to linearly separate targets from non-targets in the color model being used

(47)

Levkowitz and Herman studied the problem of creating colormaps for data visualization [51].

They found that a gray-scale (i.e., luminance-based) colormap can provide somewhere between

60 and 90 just-noticeable difference (JND) steps. Using this knowledge, they attempted to

build a linearized optimal color scale (LOCS) that offered a larger perceptual dynamic range

during visualization. They also showed that a LOCS with 32 values has a perceived color-pair

difference six times larger than a linear gray-scale colormap with 32 values [40].

Figure 4.4: Isomorphic colormap for high spatial frequency data. The high frequency colormap reveals more information in the radar data.

Figure 4.4 shows a radial sweep from a weather radar sensor measuring the high spatial

frequency variation of reflected intensity (e.g., from thick clouds). The luminance-based col-ormap being used offers a good representation of the minute details in the data [56].

Luminance is best suited for representing high spatial frequency ordinal data. Previous

work reported in [20, 21, 22] showed that a random variation of luminance may interfere with

boundary identification between two groups of differently colored elements. Callaghan sug-gests that luminance is more important than hue to the low-level visual system during boundary

(48)

4.3 Texture

Texture refers to the characteristic appearance of a surface having a tactile quality [10].

Tex-ture can be decomposed into a collection of fundamental perceptual properties such as height,

density, orientation, and regularity. Texture dimensions are usually mapped to individual

at-tributes. The result is a texture pattern that can represent multiple attribute values at a single

location.

4.3.1 Height

Height is one aspect of element size that is an important property of a texture pattern. Studies

in cognitive vision have shown that differences in height are detected preattentively by the

low-level visual system [14, 64]. Results from [49] suggest that height can support up tofive discrete values and is best suited to represent quantitative data. Furthermore, experimental

results have shown that hue and luminance cause visual interference with height [49].

4.3.2 Density

Density is an important visual feature for performing texture segmentation and classification [62]. Results from [49] suggest that density is best suited for representing low spatial frequency

ordinal data. Hue, luminance, and height cause visual interference with density [49].

4.3.3 Orientation

The visual system differentiates orientation using a collection of perceptual direction

(49)

color, density, orientation

lumi nan

ce, height,

regu larity

Figure 4.5: Groups of 3D towers are used to form a visual glyph that supports variation of the six perceptual texture dimensions (hue, luminance, height, density, orientation, and regularity)

[68]; a difference of15◦ is sufficient to rapidly distinguish elements from one another. Based on perceptual experiments, it was found that hue and luminance cause visual interference with

orientation [49].

4.3.4 Regularity

Regularity refers to the uniformity of a texture element’s spatial position, and is a visual

fea-ture that is commonly used to perform texfea-ture segmentation and classification in computer vision algorithms [62]. However, in the human visual system, differences in regularity are

difficult to detect. Typically, regularity can encode only binary information, making it best suited to represent low spatial frequency data. Hue, luminance, height, and density all cause

visual interference with regularity [49]. Figure 4.5 shows 3D glyphs that support variation of

hue, luminance, height, density, orientation, and regularity. Researchers in computer graphics

have applied texture properties such as height, density, orientation, and regularity to display

(50)

4.4 Motion

Motion is a visual feature that possesses strong perceptual cues. Motion elicits “pop-out”

effects in which moving objects can be searched in parallel by the visual system [65]. Motion

aids the process of grouping elements and is effective for providing a general overview of trends

in data [19]. The human visual system perceives, tracks, and predicts movement. Motion,

unlike color, can be rapidly detected at the periphery. Therefore, it can be used to design icons

for attracting a user’s attention toward the edge of a computer screen [16]. Different features

of motion includeflicker, direction, and velocity.

4.4.1 Flicker

Flicker refers to a repeating on-off pattern applied to an image or an object, and is normally

measured as the frequency of repetitionF in cycles per second (cps). The rate at which

succes-sive images need to be presented in order to perceive continuous motion is known as the critical

flicker frequency (CFF).F = 60 cps is often cited as the standard CFF. This number can vary depending on the color, brightness, or size of the object being displayed, and on its eccentricity

(i.e., the distance in visual angle from the viewer’s current focal point to the object). Huber and

Healey found that for rapid and accurate target detection,flicker must be coherent and have a cycle length greater than 120 milliseconds [46].

4.4.2 Direction

Direction of motion can be used in visualization to help discriminate between groups of

ele-ments with similar values. Differences in the direction that glyphs move provide cues to help

differentiate individual elements from neighboring background glyphs. Humans can

preatten-tively and simultaneously track up tofive unrelated motion trajectories within the same visual

Perceptual Display Hierarchies for Visualization

Perceptual Display Hierarchies for Visualization

Amit Prakash Sawant

Computer Science

APPROVED BY:

Contents

List of Figures

List of Tables

Chapter 1

Introduction

1.1

Perception in Visualization

1.2

Size and Visualization

1.3

Research Objectives

1.4

Thesis Organization

Chapter 2

Visual Acuity and Display Properties

2.1

Human Vision

2.1.1 Visual Angle

2.1.2 Visual Acuity

2.2

Display Device Properties

2.2.1 Display Resolution

2.2.2 Physical Size

2.2.3 Viewing Distance

Chapter 3

Visualization

3.1

Advantages of Visualization

3.2

Visualization Process

3.2.1 Visualization Pipeline

3.3

Types of Visualization

3.3.1 Scientific Visualization (SciVis)

3.3.2 Information Visualization (InfoVis)

3.4

Visualization Techniques

3.4.1 Tree-maps

3.4.2 Fisheye Lens

3.4.3 Hyperbolic Browser

Chapter 4

Visual Features

4.1

Properties of Visual Features

4.2

Color

4.2.1 Hue

4.2.2 Luminance

4.3

Texture

4.3.1 Height

4.3.2 Density

4.3.3 Orientation

4.3.4 Regularity

4.4

Motion

4.4.1 Flicker

4.4.2 Direction