perceiving objects - university of california, irvinemajumder/vispercep/chap4presentation.pdf · 2...
TRANSCRIPT
1
PERCEIVING OBJECTS
Visual Perception
Slide 2 Aditi Majumder, UCI
Different Approaches
Molecules Neurons Circuits &Brain Areas
Brain
Physiological Approach
IndividualFeatures
Groups of Features
Objects Scenes
Psychophysical Approach
2
Slide 3 Aditi Majumder, UCI
Gestalt Approach
Gestalt psychologyStructuralism : Perception is created by combining elements called sensation
But this cannot explainApparent MovementIllusory Contours
Slide 4 Aditi Majumder, UCI
Apparent Movement
3
Slide 5 Aditi Majumder, UCI
Illusory Contours
Slide 6 Aditi Majumder, UCI
Whole is different from the sum of its parts
4
Slide 7 Aditi Majumder, UCI
Basic Philosophy
The whole is different than the sum of its partsSix principles defining perceptual organization
How do we combine components to perceive the whole?Is their any basic rules that we use?
Slide 8 Aditi Majumder, UCI
Gestalt Principles of Perceptual Organization
Law of SimplicityLaw of SimilarityLaw of Good ContinuationLaw of ProximityLaw of Common FateLaw of Familiarity
5
Slide 9 Aditi Majumder, UCI
Law of Simplicity
Every stimulus pattern is seen in a way that is as simple as possible.
Slide 10 Aditi Majumder, UCI
Law of Similarity
Similar things appear to be grouped together
Shape
Lightness
Orientation
6
Slide 11 Aditi Majumder, UCI
Law of Good Continuation
Points that when connected are seen as straight or smoothly curving lines tend to be seen as belonging together, and the lines tend to be seen in such a way that they follow the smoothest path.
Slide 12 Aditi Majumder, UCI
Law of Good Continuation
Points that when connected are seen as straight or smoothly curving lines tend to be seen as belonging together, and the lines tend to be seen in such a way that they follow the smoothest path.
7
Slide 13 Aditi Majumder, UCI
Law of Proximity or Nearness
Things close together appear to be grouped togetherOvercomes law of similarity in this example
Slide 14 Aditi Majumder, UCI
Law of Common Fate
Objects moving in the same direction appear to be grouped together
8
Slide 15 Aditi Majumder, UCI
Law of Familiarity
Objects appear to be grouped if the groups appear to be familiar or meaningful.
Slide 16 Aditi Majumder, UCI
Heuristic and not Algorithm
Where does the heuristics come from?
9
Slide 17 Aditi Majumder, UCI
Palmer-Irvine Principles of Perceptual Organization
Common Region
Element Connectivity
Synchrony
Slide 18 Aditi Majumder, UCI
Quantitative Measure of Grouping Effects
Repetition Discrimination Task
10
Slide 19 Aditi Majumder, UCI
Perceptual Segregation
Gestalt TheoryReversible figure ground
Figure more object likeFigure seen as being in front of groundGround is uniform region behind figureSeparating contours appear to belong to figure
Slide 20 Aditi Majumder, UCI
Factors Determining Figure and Ground
Figure Symmetry
Smaller Areas
Horizontal or Vertical
11
Slide 21 Aditi Majumder, UCI
Factors Determining Figure and Ground
Figure Familiarity
Slide 22 Aditi Majumder, UCI
Modern Research
Role of ContoursLikeliness of Occurance
When does segregation occur?Popular belief
First segregation, then recognition
Later provedRecognition and segregation may happen in parallel
12
Slide 23 Aditi Majumder, UCI
Perceiving Objects
Molecules Neurons Circuits &Brain Areas
Brain
Physiological Approach
IndividualFeatures
Groups of Features
Objects Scenes
Psychophysical Approach
Slide 24 Aditi Majumder, UCI
How Objects are Constructed?
Marr’s computational ModelFeature Integration Theory (FIT)Recognition-by-Components Theory (RBC)
13
Slide 25 Aditi Majumder, UCI
Marr’s Theory of Object Construction
Computational Approach
Object’s imageon the retina
Identify edgesand primitives
Groups primitivesand processes
Raw primal sketch 2.5-D sketch
Perceived 3-D object
Slide 26 Aditi Majumder, UCI
Marr’s Theory
Computational ApproachCreation of raw primal sketch
Analysis of light and dark region of retinal imageUsing natural constraints
E.g. Illumination edge vs. geometric edgeDo not see this
Processed to develop a 2.5D sketch
14
Slide 27 Aditi Majumder, UCI
Feature Integration Theory (FIT)
Preattentive StageDetects features
Focused Attention StageFeatures are combined to perceive the object
Slide 28 Aditi Majumder, UCI
FIT : Preattentive Stage
Pop-out boundary for detecting featuresDifferent OrientationDifferent Value
15
Slide 29 Aditi Majumder, UCI
FIT : Preattentive Stage
Visual Search Detection TimeConstant with increase in number of distractors if target has pop-out featuresIncreases with increase in number of distractors if target has no pop-out features
Have to scan each distractor and eliminate
Similar to Salience
Slide 30 Aditi Majumder, UCI
FIT: What makes things pop out?
CurvatureTiltLine endsMovementsColorBrightnessDirection of Illumination
16
Slide 31 Aditi Majumder, UCI
FIT : Preattentive Stage
Independent featuresNo association with objectsSame observation from physiology
Slide 32 Aditi Majumder, UCI
FIT : Focused Attention Stage
Attention is essential for combining featuresSame result from physiology
17
Slide 33 Aditi Majumder, UCI
Recognition-by-Components (RBC)
Volumetric PrimitivesGeons
Principle of componential recovery
Slide 34 Aditi Majumder, UCI
Properties of Geons
View invarianceDiscriminabilityResistance to visual noise
Cannot be mathematically enforced. Not very formal. Limitation of many psychophysical model. No hard quantification.
18
Slide 35 Aditi Majumder, UCI
RBC theory
+ Can identify objects based on a few basic shapes- Cannot help us detect the finer details which causes difference
Slide 36 Aditi Majumder, UCI
Comprehensive Model
Image Based StageSurface Based StageObject Based StageCategory Based Stage
19
Slide 37 Aditi Majumder, UCI
Image Based Stage
Retinal ImageLocal feature detection
Slide 38 Aditi Majumder, UCI
Image Based Stage
Retinal ImageLocal feature detection
Edges, Corners, blobs Raw primal sketch
Global relationship between them
Full primal sketchDifficult
Similarity with Marr’s Primal Sketch
20
Slide 39 Aditi Majumder, UCI
Image Based Stage
Primitives2D structure of image intensities
Features like edges, blobs, corners etc
GeometryTwo dimensional
Reference FrameRetinal
Slide 40 Aditi Majumder, UCI
Surface Based Stage
Find intrinsic property of surfaces in the real worldSurfaces in 3D world as opposed to 2D primitivesVisible surfaces from which light reflect to our eyeIntrinsic Images
21
Slide 41 Aditi Majumder, UCI
Surface Based Stage
Represented by 2D planar elements in 3D3D surface can be represented by infinite 2D planar elementsProperties
Distance from viewerSlantShading (as color or texture)
Like a 2D rubber sheet wrapped on the face of the visible surfaces.
Similarity with Marr’s 2.5D representation
Slide 42 Aditi Majumder, UCI
Surface Based Stage
Primitives2D planes embedded in 3D
GeometryThree dimensional
Reference FrameViewer dependent
22
Slide 43 Aditi Majumder, UCI
Object Based Stage
We have some 3D definition
Otherwise, surprised when hidden surfaces got exposed
Two types of representation
2D patched in 3D3D volume elements
Hierarchical Similarity with Recognition by Components
Slide 44 Aditi Majumder, UCI
Object Based Stage
Primitives2D planes embedded in 3DVolumes
GeometryThree dimensional
Reference FrameObject dependent
23
Slide 45 Aditi Majumder, UCI
Category Based Stage
CategorizationIdentificationCognitive ScienceDeals with knowledge in perceptionHow it helps us surviveWhat about more frames? The temporal domain is not explored that much
Slide 46 Aditi Majumder, UCI
Relationship to Graphics
OpenGL triangular rendering2D triangle mesh embedded in 3DTriangle is smallest planar 2D elements
Volume RenderingUses volumes as primitives
Image based RenderingDepth Images analogous to surface based representationThat is why a view dependent rendering scheme is adoptedSince no object based information, difficulty in handling occlusion
24
Slide 47 Aditi Majumder, UCI
Role of Intelligence in Object Perception
Ambiguous StimulusInverse Projection Problem
Slide 48 Aditi Majumder, UCI
Role of Intelligence in Object Perception
Ambiguous StimulusInverse Projection Problem
Objects not separatedOcclusionAmbiguity in lightness
25
Slide 49 Aditi Majumder, UCI
Intelligence Heuristics
OcclusionLight from above