Transcript Perception - UBC Department of Computer Science

Perception
Visual Attention and Information
That Pops Out
Scales of Measurement
• Scales of Measurement
• Eye Movement
• Visual Attention, Searching, and System
Monitoring
• Reading From the Iconic Buffer
• Neural Processing, Graphemes and Tuned
Receptors
• The Gabor Model and Texture In Visualization
• Texture Coding Information
• Glyphs and Multivariate Discrete Data
Scales Of Measurement
On the Theory of Measurement, S.S. Stevens, Science, 103, pp.677-680. 1946
•
•
•
•
Nominal
Ordinal
Interval
Ratio
Nominal
• name only, arbitrary, any one-to-one
substitution allowed
• words or letters would serve as well as
numbers
• stats: number of cases, mode, contingency
correlation
• e.g numbers on sports team, names of
classes
Ordinal
• rank-ordering, order-preserving
• intervals are not assumed equal
• most measurements in Psychology use this
scale
• monotonic increasing functions
• stats: median, percentiles
• e.g. hardness of minerals, personality traits
Interval
•
•
•
•
•
quantitative, intervals are equal
no “true” zero point, therefore no ratios
Psychology aims for this scale
general linear group
stats: mean, standard deviation, rank-order
correlation, product moment correlation
• e.g. Centigrade, Fahrenheit, calendar days
Ratio
•
•
•
•
•
determination of equality of ratios (true zero)
commonly seen in physics
stats: coefficient of variation
fundamental (additivity: e.g. weights)
derived (functions of above: e.g. density,
force)
Eye Movements
• Saccadic Movement
– fixation point to fixation point
– dwell period: 200-600 msec
– saccade: 20-100 msec
• Smooth Pursuit Movement
– tracking moving objects in visual field
• Convergent Movement
– tracking objects moving away or toward us
• Saccadic suppression
– the decrease in sensitivity to visual input during
saccadic eye movement
• Brain often processing rapid sequences of
discrete images
• Accommodation
– refocusing when moving to a new target at
different distances
– neurologically coupled with convergent eye
movement
Visual Attention, Searching, and
System Monitoring
• Our visual attention is usually directed at
what we are currently fixating on.
• Supervisory Control
– complex semiautonomous systems, only
indirectly controlled by human operators
– uses searchlight metaphor
• Human-Interrupt Signal
– effective ways of computer to gain attention
• warning
• routine change of status
• patterns of events
• Visual Scanning Strategies
– Elements
• Channels, Events, Expected Costs
– Factors
• minimizing eye movement, over-sampling of
channels, dysfunctional behaviours, systematic scan
patterns
• Useful Field of View (UFOV)
– expands searchlight metaphor
– size of region from which we can rapidly take
information
– maintains constant number of targets
• Tunnel Vision and Stress
– UFOV narrows as cognitive load/stress goes up
• Role of Motion in Attracting Attention
– UFOV larger for movement detection
4 Requirements of User Interrupt
• easily perceived signal, even when outside
of area of attention
• continuously reminds user if ignored
• not too irritating
• signal conveys varying levels of urgency
How to attract user’s attention:
problems
• Difficult to detect small targets in periphery
of visual field.
• Colour blind in periphery (rods).
• Saccadic suppression allows for the
possibility of transitory events being
missed.
Movement: possible solution
• Seen in periphery.
• Research supports effectiveness of motion.
• Urgency can be effectively coded using
motion.
• Appearance of new object attracts attention
more than motion alone.
Reading from the Iconic Buffer
• Iconic Buffer
– short-lived visual buffer holds images for 1-2
seconds prior to transfer to short-term/working
memory
• Pre-attentive Processing
– theoretical mechanism underlying pop-out
– occurs prior to conscious attention
Following examples from Joanna McGrenere’s HCI class slides.
85689726984689762689764358922659865986554897689269898
02462996874026557627986489045679232769285460986772098
90834579802790759047098279085790847729087590827908754
98709856749068975786259845690243790472190790709811450
85689726984689762689764458922659865986554897689269898
85689726984689762689764358922659865986554897689269898
02462996874026557627986489045679232769285460986772098
90834579802790759047098279085790847729087590827908754
98709856749068975786259845690243790472190790709811450
85689726984689762689764458922659865986554897689269898
Pop Out
• Time taken to find target independent of
number of distracters.
• Possible indication of primitive features
extracted early in visual processing.
• Less distinct as variety of distracters
increases.
• Salience depends on strength of particular
feature and context.
Pop Out Examples
• Form:
– line orientation, length, width
– spatial orientation, added marks, numerosity (4)
• Colour:
– hue, intensity
• Motion:
– flicker, direction of motion
• Spatial Position:
– stereoscopic depth, convex/concave shape
Color
Orientation
Motion
Simple shading
Shape
Length
Width
Parallelism
Enclosure
Curvature
Added marks
Number
Spatial grouping
• Rapid Area Judgement
– fast area estimation done on basis of colour or
orientations of graphical element filling a
spatial region
• Conjunction Search
– combination of features not generally preattentive
– spatially coded information (position on XY
plane, stereoscopic depth, shape from shading)
and second attribute (colour, shape) DO allow
conjunction search
Neural Processing, Graphemes,
and Tuned Receptors
• Cells in Visual Areas 1 and 2 differently
tuned to:
–
–
–
–
orientation and size (with luminance)
colour (two types of signal)
stereoscopic depth
motion
• Massively parallel system with tuned filters
for each point in visual field.
Vision Pathway
http://www.geocities.com/ocular_times/vpath2.html
• Signal leaves retina, passes up optic
nerve, through neural junction at
geniculate nucleus (LGN), on to
cortex.
• First areas are Visual Area 1 and
Visual Area 2: these areas have
neurons with preferred orientation and
size sensitivity (not sensitive to
colour)
http://www.geocities.com/ocular_times/vpath.html
http://www.geocities.com/ocular_times/vpath.html
http://nba5.med.uth.tmc.edu/academic/neuroscience/lectures/section_2/lecture34_04.htm
http://nba5.med.uth.tmc.edu/academic/neuroscience/lectures/section_2/lecture34_04.htm
Grapheme
• Smallest primitive elements in visual
processing, analogous to phonemes.
• Corresponds to pattern that the neuron is
tuned to detect (‘filter’).
• Assumption: rate of neuron firing key
coding variable in human perception.
Gabor Model and Texture in
Visualization
• Mathematical model used to describe
receptive field properties of the neurons of
visual area 1 and 2.
• Explains things in low-level perception:
– detection of contours at object boundaries
– detection of regions with different visual
textures
– stereoscopic vision
– motion perception
Gabor Function
•
•
•
•
Response = C cos(Ox/S)exp(-(x² + y²)/S)
C amplitude, or contrast value
S overall size of Gabor function
O rotation matrix that orients cosine wave
• orientation, size, and contrast are most
significant in modeling human visual
processing
• Gabor model helps us understand how the
visual system segments the visual world
into different textual regions.
• Regions are divided according to
predominant spatial frequency(grain or
coarseness of a region)
and orientation information
• Regions of an image are analyzed
simultaneously with Gabor filters, texture
boundaries are detected when best-fit filters
for one area are substantially different from
a neighbouring area.
Trade-Offs in Information
Density
• The second dogma (Barlow, 1972)
– visual system is simultaneously optimized in
both spatial-location and spatial-frequency
domains
• Gabor detector tuned to specific orientation
and size information in space.
• Orientation or size can be specified exactly,
but not both, hence the trade-off.
Texture Coding Information
• Gabor model can be used to produce easily
distinguished textures for information
display (used to represent continuous data).
• Human neural receptive fields couple the
gaussian and cosine components, resulting
in three parameter model:
– O orientation
– S scale / size
– C contrast / amplitude
• Textons
– combinations of features making up small
graphical shapes
• Perceptual Independence
– independence of different sources of
information, increase in one does not effect
how the other appears
• Orthogonality
– channels that are independent are orthogonal
– textures differing in orientation by +/- 30
degrees are easily distinguishable
Texture Resolution
• Resolvable size difference of a Gabor
pattern is 9%.
• Resolvable orientation difference is 5°.
• Higher sensitivity due to higher-level
mechanisms.
• No agreement on what makes up important
higher order perceptual dimensions of
texture (randomness is one example).
Glyphs and Multivariate Discrete
Data
• Multivariate Discrete Data
– data objects with a number of attributes that can
take different discrete values
• Glyph
– single graphical object that represents a
multivariate data object
• Integral dimensions
– two or more attributes of an object are
perceived holistically (e.g.width and height of
rectangle).
• Separable dimensions
– judged separately, or through analytic
processing (e.g. diameter and colour of ball).
• Restricted Classification Tasks
– Subjects asked to group 2 of 3 glyphs together
to test integral vs. separable dimensions.
• Speeded Classification Tasks
– Subjects asked to rapidly classify glyphs
according to only one of the visual attributes to
test for interference.
• Integral-Separable Dimension Pairs
– continuum of pairs of features that differ in the
extent of the integral-separable quality
– integral(x/y size)…separable(location/colour)
Multidimensional Discrete Data
• Using glyph display, a decision must be
made on the mapping of the data dimension
to the graphical attribute of the glyph.
• Many display dimensions are not
independent (8 is probably maximum).
• Limited number of resolvable steps on each
dimension (e.g. 4 size steps, 8 colours..).
• About 32 rapidly distinguishable
alternatives, given limitations of
conjunction searches.
Conclusion
• What is currently known about visual
processing can be very helpful in
information visualization.
• Understanding low-level mechanisms of the
visual processing system and using that
knowledge can result in improved displays.