Visual cognition theories Feature detection in object recognition Denis...

Object recognition 3

Visual cognition theories

Feature detection in object recognition

Denis PelliPsychology and Neural Science, NYU

December 10, 2007

Outline1. Critical comments on existing theories, in visual cognition, for how peoplerecognize objects.

2. Features

Outline1. Critical comments on existing theories, in visual cognition, for how peoplerecognize objects.Treisman & Kanwisher 1998 Perceiving visually presented objects: recognition, awareness, andmodularity. Current Opinion in Neurobiology.

2. Features

Treisman & Kanwisher 1998 Perceiving visually presented objects: recognition, awareness, andmodularity. Current Opinion in Neurobiology.

How do we recognize objects? What are the theories? The centralcontroversies in the field have been dichotomies:Parts vs. wholeStructural vs. viewpoint-dependentConscious vs. unconsciousPerception vs. actionDorsal vs. ventral

Lots of papers have accumulated evidence bearing on these issues, showingdifferences along these scales between tasks. In some cases patients revealdissociations. These dimensions are real; there are data to be explained.

However, while initial positions were at either ends of the dichotomies,everyone has since drifted to moderate views that allow for intermediatepositions or combinations of both extremes. As a consequence, no one iswrong. Since no one is wrong, one may well wonder whether the intermediatepositions that embrace the whole gamut are testable scientific theories. Canthey be refuted?

Except for part/whole in face recognition, these theories haven’t helped much inexplaining everyday object recognition.


Priming, matching, and repetition blindness are all object-specific, yet invariantacross views.

Object files (Pylyshyn, multiple object tracking)

Pylyshyn, multiple object tracking


“Wolfe (1998) has collected surprising evidence that previously attended objecttokens revert to a similar unstructured state once attention is withdrawn,concluding that, ‘Vision exists in the present tense. It remembers nothing.’”


Modularity

Modularity vs. consciousness

There is much evidence of modularity in the brain, in which someareas seem to know things that other areas don’t. Viewing thebrain as a machine, this is an old familiar result from thenineteenth century, and a perfectly reasonable way to build abrain. However, if we are talking about the human brain, then themodularity challenges our intuitions about consciousness.

Ned Block distinguishes between phenomenal consciousness andaccess consciousness. Phenomenal consciousness consists ofsubjective experience and feelings. Access consciousnessconsists of that information globally available in the cognitivesystem for the purposes of reasoning, speech and high-levelaction control. The key word is “globally”. Our intuition is thatinformation should never be inconsistent among different parts ofthe same brain, or different aspects of our behavior.

Outline1. Critical comments on existing theories, in visual cognition, for how peoplerecognize objects.

2. FeaturesSame features for letter identification as for grating detection.Features do not scale with letter size.Features are detected independently.

Solomon & Pelli (1994) Nature

Solomon & Pelli (1994) Nature

Gain

c/letter

0.15

0.6

2.4

9.7

Majaj, Liang, Martelli, Berger & Pelli (2003) Vision Sciences Society

Majaj, Pelli, Kurshan, & Palomares (2002) Vision Research

Robson & Graham (1981) Vision Research

English (Bookman) a b c d e f g h i j k l m n o p q r s t u v w x y z

English (uppercase Bookman) ABCDEFGHIJKLMNOPQRSTUVWXYZ

English (bold Bookman) a b c d e f g h i j k l m n o p q r s t u v w x y z

English (Courier) abcdefghijklmnopqrstuvwxyz

English (Helvetica) a b c d e f g h i j k l m n o p q r s t u vw x y zEnglish (Künstler) A B C D E F G H I J K L M N O P Q R S T U … Z

Sloan C D H K N O R S V Z

Arabic x , i K LG : R T YU 3 4 Q W A S . v 5 6 OP { E J F hArmenian abgd;xhejvil.`kf][ym\n,ocp=®swtrz'q

Chinese abcdefghijklmnopqrstuvwxyzDevanagari a b c d E f g h D j k l m n Z p q r s t A v W x y z

Hebrew t c r q x p [ s n m l k y f j z w h d g b a2! 3 Checkers a b c d e f g h i j k l m n o p q r s t u v w x y z

4! 4 Checkers abcdefghijklmnopqrstuvwxyz

3-letter words all and any are but can for had has her him his its … you

5-letter words about after being could first great house might never … years

0.01

0.1

1

10 100

Figure 4a

Effi

cien

cy

Complexity

d

a

aK afman Aa

awas

a

whichA

a

Pelli, Burns, Farell, & Moore (2005) Vision Research

Efficiency

perimeter2

“ink” areaComplexity

Pelli, Farell, & Moore (2003) Nature

0.01

0.1

1 10

Efficiency

Word!length!(letters)

b


10

100

1000

1 10

Ener

gy (r

e no

ise)

Word!length!(letters)

a

Ideal

Human


Of course you can read letter by letter,

but can you read word by word?


Outline1. Critical comments on existing theories, in visual cognition, for how peoplerecognize objects.Treisman & Kanwisher 1998 Perceiving visually presented objects: recognition, awareness, andmodularity. Current Opinion in Neurobiology.

2. FeaturesSame features for letter identification as for grating detection.Features do not scale with letter size.Features are detected independently.

Visual cognition theories Feature detection in object recognition Denis...

Documents

Transcript of Visual cognition theories Feature detection in object recognition Denis...