Lecture 14 - object perception and recognition.

Question 1

how do we extract info from a scene?

Answer 1

we use a number of cues in the environment to help us along (top down)

global features and cues within complex environments

Question 2

global image features

Answer 2

– Degree of naturalness (how green, how rounded)

– Degree of openness

– Degree of roughness (complexity)

– Degree of expansion (looking for depth cues: things further away or closer )

– Color

holistic and rapid

critical: not hard-wired, we do not come with them: we learn them over time (that’s what makes them top down)

Question 3

Physical regularities

Answer 3

regularly occurring physical properties and that we can learn about

• oblique effect
• uniform connectedness

Question 4

oblique effect

Answer 4

really good at finding vertical and horizontal arrangements in an environment

people perceive horizontals and
vertical more easily than other orientations.

Question 5

uniform connectedness

Answer 5

anything that seems connected, by color, texture, we pick out immediately

objects are defined by
areas of the same color or texture.

Question 6

light from above heuristic

Answer 6

purely based on experience: light in natural
environment comes from above us.

we carry with us an expectation that light is coming from above

perceive things differently based on where shadows are

Question 7

semantic regularities

Answer 7

things associated with what we expect to happen

characteristics associated with the
functions of scenes

Question 8

Stephen Palmer’s (1975) semantics experiment

Answer 8

given a context scene, and then very briefly you get one target image (loaf of bread, mailbox, drum) where one of them is semantically congruent (a loaf of bread in a kitchen) because it is what we would expect to see in this context

– Results showed that:

• Targets congruent with the context were identified 80% of the time .
• Targets that were incongruent were only identified 40% of the time.

strong priming effect: more quickly and accurately recognize objects in a scene

Question 9

Theory of unconscious inference

Answer 9

ambiguous or incompleting info will be interpreted in congruent ways (educated guesses about what we expect to see)

your perceptions were a result of some unconscious assumptions (learned expectations: system primed based on experience)

– Created by Helmholtz (1866/1911) to explain why ambiguous stimuli can be interpreted in specific ways.

– Main Principle: perceptions are result of unconscious assumptions about the environment.

– Likelihood principle -

Question 10

Likelihood principle

Answer 10

objects are perceived based on what is most likely to have caused the pattern.

Question 11

Bayesian inference

Answer 11

taking prior probabilities into account anytime you’re making a guess about the environment

based on what’s occurred in the past, what’s likely to happen?

Modern researchers take on Theory of unconscious inference

Question 12

to help us make inferences about the

environment we use

Answer 12

learned regularities

Question 13

where in the brain is this happenening?

Answer 13

Neurons in striate visual cortex
respond to Gestalt grouping
principles.

Question 14

Contextual modulation -

Answer 14

• stimuli outside of a neuron’s (classical)
  receptive field can affect neural firing rate

– Happens when these stimuli follow good continuation: stimuli outside the receptive field are following heuristic characteristics

– Happens when the stimuli are perceived as part of the figure.

– Consistent with the role of top-down
influence on perception and recognition.

• isn’t feedforward or lateral inhibition, it’s topdown info coming from visual cortex

Question 15

Contextual modulation
(cont.)

• A static bar

Answer 15

• A static bar activates cells if presented with a congruent background (object-like background), causing activation in V1.

• Remove the congruent background, but keep the same bar, and the activation stops. [No
longer part of square that was different from the entire scene.] - changed whether or not it belongs to an object (the thing you would manipulate within a scene)

• How does the cell know? May have something to do with processing later that provides feedback to the cell.

shows that really early on cells are responding !!

Question 16

The use of single cell recordings in monkey V1, which found
cells that seem to respond to the Gestalt principle of good
continuation, is an example of the _________________ level
supporting the __________________ level of analysis.

Answer 16

Implementational, algorithmic

Question 17

Different frameworks of object perception (domain specificity & property based) make

Answer 17

different predictions
about neural implementation [sparse vs. population
codes].

Question 18

Perceptual organization biases can be described by

Answer 18

gestalt principles (it’s a model that doesn’t tell you what the implementation should be)

Question 19

Recognition-by-components (RBC) theory can be added to gestalt principles to account for

Answer 19

phenomena like viewpoint invariance and occluded objects.

Question 20

We use a number of cues [physical and semantic] to extract information from

Answer 20

the environment (scenes).