What is the inverse problem?
Any retinal image is consistent with infinitely many possible configurations of the wolrd
What are the multiple 3D cues?
- binocular
- motion
- pictoral
- oculomotor
What is binocular disparity (stereo vision)?
It arises because we have two eyes in slightly different positions in our head giving us a slightly different view of the world so our retinal image is slightly different on each eye. We use these differences (binocular disparities) to see the world
What are the two types of motion cues?
- Motion parallax: due to self-motion. Any time we make a movement thing on a retina move. Things that are close to use move more and things that are further away move less. The visual system can use the speed of something moving to work out how far away from us it is
- Kinetic depth (KDE): due to object motion. Different amounts of the surface will move by different amounts. Visual system can use differences in how different bits of the surface are moving to figure out what the object was.
What are pictorial cues?
when texture elements change in their size
What is elevation pictorial cue?
things lower down in the image tend to be closer
What is the relative pictorial cue?
things than are bigger tend to be closer
What is the perspective pictorial cue?
lines tend to converge
What is the shading pictorial cue?
Patterns of light and dark give impression of concavity
What is occlusion?
Occluding objects show what is closer and further away
What are the two types of oculomotor cues?
Convergence and Accommodation?
What is convergence?
- It arises because our eyes have to bend towards each other in order to fixate the same object
- they will bend more to view objects that are close to you vice versa
- muscles that cause eyes to converge will send signals to the visual system showing how far away an object is
What is accommodation?
The lens of the eye changes in response to how far away the object is. The ciliary muscles send signals to the visual system to tell how far away the object is
What is the problem with depth perception if we have lots of different cues?
- Many cues are ambiguous - 2D image is compatible with infinite 3D worlds
- With multiple cues available - how do we perceive a single unified world?
How can we overcome ambiguity in 3D cues?
By using prior knowledge to interpret the image. These assumptions are gained through our knowledge and experience of the physical properties of the world. This is a type of top down processing and supports the constructivist approach to vision
What assumption do we make to overcome ambiguity in perspective?
We make the assumption that lines in the world tend to be parallel
What assumption do we make to overcome ambiguity in shading?
We assume that light comes from above
What assumptions do we make to overcome ambiguity in texture and what does making this assumption mean?
- we assume surface textures are isotropic (unbiased orientation - all orientation is the same) and homogenous (uniform density)
- making this assumption means any changes in image texture orientation or density are attributed to changes in 3D surface orientation
What assumption do we make to overcome ambiguity in elevation?
We assume that objects rest on a ground plane (because we have grown up in a world where we experience gravity)
What do invalid assumptions lead to?
Illusions
What does integration of cues help overcome problems of?
- reliability (some cues are more reliable than others)
- ambiguity (some cues are ambiguous)
- Conflict (sometimes cues give us different estimates of depth)
What are the three types of multi cue integration?
- Compromise
- Dominance
- Interaction
What is the compromise multi cue integration?
when two sources of depth information are conflicting the brain tries to find a compromise between the two. The brain tries to average them but the final percept of shape will be biased towards the most reliable cue
What is the dominance multi cue integration?
when two cues define very different shapes or depths the brain may choose to ignore one in preference for another
