Chapter 6 Object Perception. Object recognition 0.1 sec to identify many objects. We typically don’t encounter things tachistoscopically; but we instead.

Chapter 6

Object Perception

Object recognition

0.1 sec to identify many objects. We typically don’t encounter things

tachistoscopically; but we instead scan. Fixations last typically about .25 sec.

Challenges in object recognition

Occlusion (desk, screen) Upside down Picture of a chair vs. chair Far chair in room vs. toy chair on desk

Two types of theories for object recognition Recognition by components (Biederman) View-based recognition (Bülthoff, Tarr, et

al.)

Geons:

Distinguishable from almost any perspective.

Recognizable even with occlusions.

Pros and cons of geon theory

Explains why it is hard to recognize objects from unusual angles.

No physiological evidence. Sometimes recognition is viewpoint

dependent. Doesn’t explain recognition of individuals.

View-based recognition

People have preferred viewing angles for familiar objects.

Less consensus on preferred viewing angles for novel objects.

Recognition is slower the greater the deviance from preferred viewing angle.

Impact of rotation on recognition

James Stone (1998) showed ameboid images.

Impact of rotation on recognition

James Stone (1998) showed ameboid images.

Guy Wallis (1998) five head shots in apparent rotation.

Learning to see

The young woman – old woman illusion James Elkins experiences with moth

hunting Face - woman illusion High contrast images

Perceptual learning

(p. 230) Panel A: Orientation discrimination Panel B: Vernier acuity (The degree to which a

pair of fine lines can be aligned to each other.) Panel C: Orientation of 3-line bars (followed by

mask) Panel D: Face recognition in the presence of

noise.

Perceptual learning

Chicken sexing Perception of high contrast images (p.

232)Pre-training: 55% for faces; 13% for others.Post-training: 93% for faces; 87% for others.

Inferotemporal cortex

Single cell recording reveals:Few neurons that prefer oriented lines.Neurons with large receptive fields.Neurons that are sensitive to “diagnostic”

features, such as eyes of a face.Neurons that are occlusion insensitive.Neurons that are object size invariant.

Dolan: Viewing gray-scale images facilitates interpreting high contrast versions of those images.

Tovee, Rolls, and Ramachandran, (1996). Monkeys shown black & white, then gray-scale, then black & white again. IT cells were more active the second time around.

Kobatake, Wang, & Tanaka, (1998).

Monkeys shown a target, then after a delay the target with three other objects.

Monkeys improved with practice to be able to perform with longer delays.

IT cells changed firing patterns.

Face Recognition

Much worse when contrast is reversed. Much worse when inverted. The top half of a face is harder to

recognize, when the face is hybridized. Anger is the most distinctive of the six basic

facial expressions (sadness, happiness, anger, fear, surprise, and disgust).

Prosopagnosia

Two types of strategies to cope:Configural: patients base their judgements on

the overall similarity of the faces’ configuration.

Featural processing: patients base their judgements on parts of the face, rather than the whole.

Greebles

http://mail.cog.brown.edu/~tarr/projects/greeble_poster.html

They have symmetric body parts, families, and genders.

Becoming a greeble expert.

Greebles

Expert recognition: As good with general categories as with specific categories.Bird and black-capped chickadeeSex, family, and individual identity

People can become greeble experts after thousands of trials.

Become a Greeble expert

The better a person is at greeble recognition, the more than fMRIs look like facial fMRIs.

The ‘face’ area appears to be deployed in the inspection of even novel greebles.

Attention and object recognition

Attention reduces variability in perception.Prinzmetal, et al., show that distractors

increase trial-to-trial variability in color assessments.

Attention changes an object’s appearance Attention can increase sensitivity of

appearances.

Spotlight versus object attention

Edgly, Driver, and Rafal (1994)

Attention and object recognition

Egly, Driver, and Rafal (1994) suggest that attention is object-based.

The gorilla and girl with umbrella examples also suggest that vision is not a spotlight, but is object-based.

Change Blindness

See demos from Rensink at University of South Dakota.

Imagery and Vision

Cheves Perky (1910). Confusability of imagination and vision

fMRI has shown that some of the same brain regions are active during both vision and imagining, including V1. (Kosslyn, et al., 1999)

Imagery and Vision

Cheves Perky (1910). Confusability of imagination and vision

fMRI has shown that some of the same brain regions are active during both vision and imagining, including V1. (Kosslyn, et al., 1999)

TMS to V1 shows that response times on this task are slowed by ~200msec. About the same for both imagery and vision.

What does visual imagery look like? Compare doctored pictures with mental

images. Rate them for similarity on a scale 1-7 when

simultaneously present. Rate them for similarity on a scale 1-7 when

successively present. Images look like vision, except with high

frequency detail removed.

Perceptual Aspects of Reading

Alexia = inability to read acquired late in life.Letter by letter alexia appears to be due to a

visual processing deficit.

The number of saccades and fixation times depends on reading ability and difficulty of material.

Basic facts on reading saccades

Saccades occur on average 0.25 sec. Range of 0.1-0.5 sec. Distance of 2 – 18 letters. 90% of saccades among English readers

are left to right. By making text fill one spot, it is possible

to increase reading rates 3-4 times.

Word superiority effect

Letters can be better recognized in the context of words, than in isolation.

Thus, it appears that words are processed holistically, rather than individually.

Reading through a window

One letter at a time: 90% accuracy on words.

7-9 letters (so entire words could be seen), words were read error-free, but more slowly than normal.

Four words at a time, reading was still slower than normal.

Contextual influences

Expectations shape reading. Unexpected words involve longer fixations.

Spaces in reading

Spaces are not necessary for reading. In fact, you can learn to read about as accurately and with comprehension with as without them.

Fixation points were largely unchanged with the removal of spaces.

Insertion of spaces in the wrong place lead to slowed processing.

Spaces and meaning

Booth, Epelboim, and Steinman (1996). Meaning versus interword spaces.

MeaningfulMeaningless

Normal spacing 100% <<100%

No spacing <100% <<100%

Temporary failures of processing What happens in the face of extreme

ambiguity? Bruner and Potter (1964), studied

perception of defocused images. If you start with more defocused images,

recognition is slowed.Hypothesis: People always conjecture

something and then persevere.

Age and recognition of defocused images Adults are 50% faster than children in

recognition. Adults shape their guesses in a

continuous way; children change their guesses randomly at each stage.