i
i
i
i
i
i
i
i
2 2
2 2
William B. Thompson
Visual Perception
The ultimate purpose of computer graphics is to produce images for viewing by
people. Thus, the success of a computer graphics system depends on how well it
conveys relevant information to a human observer. The intrinsic complexity of the
physical world and the limitations of display devices make it impossible to present
a viewer with the identical patterns of light that would occur when looking at a
natural environment. When the goal of a computer graphics system is physical
realism, the best we can hope for is that the system be perceptually effective:
displayed images should “look” as intended. For applications such as technical
illustration, it is often desirable to visually highlight relevant information and
perceptual effectiveness becomes an explicit requirement.
Artists and illustrators have developed empirically a broad range of tools and
techniques for effectively conveying visual information. One approach to improv-
ing the perceptual effectiveness of computer graphics is to utilize these methods
in our automated systems. A second approach builds directly on knowledge of
the human vision system by using perceptual effectiveness as an optimization cri-
teria in the design of computer graphics systems. These two approaches are not
completely distinct. Indeed, one of the first systematic examinations of visual
perception is found in the notebooks of Leonardo da Vinci.
The remainder of this chapter provides a partial overview of what is known
about visual perception in people. The emphasis is on aspects of human vision
that are most relevant to computer graphics. The human visual system is ex-
tremely complex in both its operation and its architecture. A chapter such as this
553
i
i
i
i
i
i
i
i
554 22. Visual Perception
can at best provide a summary of key points, and it is important to avoid over
generalizing from what is presented here. More in-depth treatments of visual per-
ception can be found in Wandell (1995) and Palmer (1999); Gregory (1997) and
Yantis (2000) provide additional useful information. A good computer vision ref-
erence such as Forsyth and Ponce (2002) is also helpful. It is important to note
that despite over 150 years of intensive research, our knowledge of many aspects
of vision is still very limited and imperfect.
22.1 Vision Science
Vision is generally agreed to be the most powerful of the senses in humans.
Vision produces more useful information about the world than does hearing,
Light:
• travels far
• travels fast
• travels in straight lines
• interacts with stuff
• bounces off things
• is produced in nature
• has lots of energy
—Steven Shafer
Figure 22.1. The nature of
light makes vision a power-
ful sense.
touch, smell, or taste. This is a direct consequence of the physics of light (Fig-
ure 22.1). Illumination is pervasive, especially during the day but also at night
due to moonlight, starlight, and artificial sources. Surfaces reflect a substantial
portion of incident illumination and do so in ways that are idiosyncratic to par-
ticular materials and that are dependent on the shape of the surface. The fact
that light (mostly) travels in straight lines through the air allows vision to acquire
information from distant locations.
The study of vision has a long and rich history. Much of what we know
about the eye traces back to the work of philosophers and physicists in the 1600s.
Starting in the mid-1800s, there was an explosion of work by perceptual psy-
chologists exploring the phenomenology of vision and proposing models of how
vision might work. The mid-1900s saw the start of modern neuroscience, which
investigates both the fine-scale workings of individual neurons and the large-scale
architectural organization of the brain and nervous system. A substantial portion
of neuroscience research has focused on vision. More recently, computer science
has contributed to the understanding of visual perception by providing tools for
precisely describing hypothesized models of visual computations and by allow-
ing empirical examination of computer vision programs. The term vision science
was coined to refer to the multidisciplinary study of visual perception involving
perceptual psychology, neuroscience, and computational analysis.
Vision science views the purpose of vision as producing information about
objects, locations, and events in the world from imaged patterns of light reach-
ing the viewer. Psychologists use the term distal stimulus to refer to the physical
world under observation and proximal stimulus to refer to the retinal image.
1
Us-
1
In computer vision, the term scene is often used to refer to the external world, while the term
image is used to refer to the projection of the scene onto a sensing plane.
i
i
i
i
i
i
i
i
22.2. Visual Sensitivity 555
ing this terminology, the function of vision is to generate a description of aspects
of the distal stimulus given the proximal stimulus. Visual perception is said to be
veridical when the description that is produced accurately reflects the real world.
In practice, it makes little sense to think of these descriptions of objects, locations,
and events in isolation. Rather, vision is better understood in the context of the
motor and cognitive functions that it serves.
22.2 Visual Sensitivity
Vision systems create descriptions of the visual environment based on properties
of the incident illumination. As a result, it is important to understand what prop-
erties of incident illumination the human vision system can actually detect. One
critical observation about the human vision system is that it is primarily sensi-
tive to patterns of light rather than being sensitive to the absolute magnitude of
light energy. The eye does not operate as a photometer. Instead, it detects spatial,
temporal, and spectral patterns in the light imaged on the retina and information
about these patterns of light form the basis for all of visual perception.
There is a clear ecological utility to the vision system’s sensitivity to variations
in illumination over space and time. Being able to accurately sense changes in the
environment is crucial to our survival.
2
A system which measures changes in
light energy rather than the magnitude of the energy itself also makes engineering
sense, since it makes it easier to detect patterns of light over large ranges in light
intensity. It is a good thing for computer graphics that vision operates in this
manner. Display devices are physically limited in their ability to project light
with the power and dynamic range typical of natural scenes. Graphical displays
would not be effective if they needed to produce the identical patterns of light as
the corresponding physical world. Fortunately, all that is required is that displays
be able to produce similar patterns of spatial and temporal change to the real
world.
22.2.1 Brightness and Contrast
In bright light, the human visual system is capable of distinguishing gratings con-
sisting of high contrast parallel light and dark bars as fine as 50–60 cycles/degree.
(In this case, a “cycle” consists of an adjacent pair of light and dark bars.) For
2
It is sometime said that the primary goals of vision are to support eating, avoiding being eaten,
reproduction, and avoidance of catastrophe while moving. Thinking about vision as a goal-directed
activity is often useful, but needs to be done so at a more detailed level.
i
i
i
i
i
i
i
i
556 22. Visual Perception
Figure 22.2. The contrast between stripes increases in a constant manner from top to
bottom, yet the threshold of visibility varies with frequency.
comparison, the best currently available LCD computer monitor, at a normal
viewing distance, can display patterns as fine as about 20 cycles/degree. The
minimum contrast difference at an edge detectable by the human visual system
in bright light is about 1% of the average luminance across the edge. In most
8-bit displays, differences of a single gray level are often noticeable over at least
a portion of the range of intensities due to the nature of the mapping from gray
levels to actual display luminance.
Characterizing the ability of the visual system to detect fine scale patterns (vi-
sual acuity) and to detect changes in brightness is considerably more complicated
than for cameras and similar image acquisition devices. As shown in Figure 22.2,
there is an interaction between contrast and acuity in human vision. In the figure,
the scale of the pattern decreases from left to right while the contrast increases
from top to bottom. If you view the figure at a normal viewing distance, it will
be clear that the lowest contrast at which a pattern is visible is a function of the
spatial frequency of the pattern.
There is a linear relationship between the intensity of light L reaching the eye
from a particular surface point in the world, the intensity of light I illuminating
that surface point, and the reflectivity R of the surface at the point being observed:
L = αI ·R, (22.1)
i
i
i
i
i
i
i
i
22.2. Visual Sensitivity 557
Figure 22.3.
Lightness constancy
. Cast a shadow over one of the patterns with your hand
and notice that the apparent brightness of the two center squares remains nearly the same.
where α is dependent on the relationship between the surface geometry, the pat-
tern of incident illumination, and the viewing direction. While the eye is only
able to directly measure L, human vision is much better at estimating R than L.
To see this, view Figure 22.3 in bright direct light. Use your hand to shadow one
of the patterns, leaving the other directly illuminated. While the light reflected off
of the two patterns will be significantly different, the apparent brightness of the
two center squares will seem nearly the same. The term lightness is often used
to describe the apparent brightness of a surface, as distinct from its actual lumi-
nance. In many situations, lightness is invariant to large changes in illumination,
a phenomenon referred to as lightness constancy.
The mechanisms by which the human visual system achieves lightness con-
stancy are not well understood. As shown in Figure 22.2, the vision system is
relatively insensitive to slowly varying patterns of light, which may serve to dis-
count the effects of slowly varying illumination. Apparent brightness is affected
by the brightness of surrounding regions (Figure 22.4). This can aid lightness
constancy when regions are illuminated dissimilarly. While this simultaneous
contrast effect is often described as a modification of the perceived lightness of
(a) (b)
Figure 22.4. (a) Simultaneous contrast: the apparent brightness of the center bar is affected
by the brightness of the surrounding area; (b) The same bar without a variable surround.
..................Content has been hidden....................
You can't read the all page of ebook, please click here login for view all page.