Lecture 05 - Object Recognition
🧠
The challenge of object recognition:linking currentexperience w/ past experience(memory)
Theories
🧠
template theory:recognizing objects as a whole ratherthan breaking it down into the parts that
make it up
○ template: a memorized notion of something you’ve seenbefore
○ when you see an object you align it to the template
○ consistencyis key for this theory
○ requires you to havea template for every single thingyou’ve seen in life, in every possible
orientation/angle you’ve seen it in
■ that’s a lot of fucking brain space
■ as we learn, new templates are created
○ barcode anecdote
■ a particular barcode is associated w/ a particular item in the entire store, likewise a
particular item in a particular orientation matches up with one of the million templates
in your mind
○ the problemw/ this theory is that itassumes thebrain is 10x largerthan it actually is,cause
🧠CAPTCHAexplained throughtemplate theory
○ bots have difficulty computing the CAPTCHA tasks because they don’t align with any
templates the technology has
○ gatekeepers bots from buying T-Swift tickets (except they’re becoming more advanced and
learning from humans completing the CAPTCHAs to bypass it
🧠feature analysis theories:recognizing an object bybreaking it down into the partsthat make it up
○ an object is made up of a particular arrangement of features [lines, shapes, etc.]
➔ The “T” is comprised of particular featuresthat
are stored in our minds as mental representations,
when we notice those features together we realize
we’re looking at a “T”
🧠recognition by components (RBC):recognizing an objectby its 3D components (referred to as
geons)
○ think of it as the“lego theory”,geons = buildingblocks that can be combined to make sets
○ states we have various geons in our minds thatdifferin terms of shape[contour, number of
connection points, symmetry, size, etc.]
○ differs from feature analysis theoryas it looks ata 3D shape rather than the features that would
make up a shape [i.e. lines]
, ○ differs from template theorybecause you don’t need to have a geon in diff angles to recognize
the objects
■ Biederman (theory developer) states, however, that you would need to see all the geons
of the object in order to recognize it
○ alimitationto the theory is that it accounts forshape and orientation of the geons, but
doesn’t necessarily account for other physical properties that can differentiate objects from
each other [texture, colour, etc.]
🧠all theories have different viewpoints on whether or not the perspective in which the object is viewed
matters
○ view-dependent:orientationmattersin order to recognizean object
■ i.e.template theory
○ view-invariant:orientationdoes notmatterin orderto recognize an object
■ i.e.RBC theory
🧠repetition suppression (RS):brain activity decreaseswhen your brain gets used to looking at a
stimuli
○ repetition:repeated exposure
○ suppression:reduction in brain activity
○ this phenomenon was discovered shortly after the development of fMRI
○ to observe this phenomenon, you would use the subtraction method
■ [neural activity for 1st presentation] - [neural activity for 2nd presentation]
● present a stimulus for the first time, note the brain areas that activate [i.e.
attention, visual perception, etc.]
● take the stimulus away for a bit
● present the stimulus again and take note on the activated brain areas
○ you’ll notice that there would be less areas active since your brain is
getting used to the stimulus and has a better mental picture of it
○ this reduction in brain activitydoesn’t hinder recognitionability, rather is more about the
brain beingmore efficientwhen it recognizes objects
○ hypothesis: each time you see a stimulus the brainfine tunes the features that make up the
object, filters out unnecessary features [i.e. the colour of an umbrella not an important
defining feature of umbrella]
○ this theory can work for any visual stimulus [i.e. faces]
■ thebrain recalibrates when there’s a dramatic change[Kaaviya w/ straight blue hair]
🧠the fusiform area located in the temporal lobe is known for object recognition
🧠using RS to test out if perspective of an object’s orientation matters
○ fusiform area in the left hemisphere = view-invariant
○ fusiform area in the right hemisphere = view-dependent
○ ergo, the brain isboth view-invariant and view-dependent,depending on the side of brain
you’re looking at
🧠
The challenge of object recognition:linking currentexperience w/ past experience(memory)
Theories
🧠
template theory:recognizing objects as a whole ratherthan breaking it down into the parts that
make it up
○ template: a memorized notion of something you’ve seenbefore
○ when you see an object you align it to the template
○ consistencyis key for this theory
○ requires you to havea template for every single thingyou’ve seen in life, in every possible
orientation/angle you’ve seen it in
■ that’s a lot of fucking brain space
■ as we learn, new templates are created
○ barcode anecdote
■ a particular barcode is associated w/ a particular item in the entire store, likewise a
particular item in a particular orientation matches up with one of the million templates
in your mind
○ the problemw/ this theory is that itassumes thebrain is 10x largerthan it actually is,cause
🧠CAPTCHAexplained throughtemplate theory
○ bots have difficulty computing the CAPTCHA tasks because they don’t align with any
templates the technology has
○ gatekeepers bots from buying T-Swift tickets (except they’re becoming more advanced and
learning from humans completing the CAPTCHAs to bypass it
🧠feature analysis theories:recognizing an object bybreaking it down into the partsthat make it up
○ an object is made up of a particular arrangement of features [lines, shapes, etc.]
➔ The “T” is comprised of particular featuresthat
are stored in our minds as mental representations,
when we notice those features together we realize
we’re looking at a “T”
🧠recognition by components (RBC):recognizing an objectby its 3D components (referred to as
geons)
○ think of it as the“lego theory”,geons = buildingblocks that can be combined to make sets
○ states we have various geons in our minds thatdifferin terms of shape[contour, number of
connection points, symmetry, size, etc.]
○ differs from feature analysis theoryas it looks ata 3D shape rather than the features that would
make up a shape [i.e. lines]
, ○ differs from template theorybecause you don’t need to have a geon in diff angles to recognize
the objects
■ Biederman (theory developer) states, however, that you would need to see all the geons
of the object in order to recognize it
○ alimitationto the theory is that it accounts forshape and orientation of the geons, but
doesn’t necessarily account for other physical properties that can differentiate objects from
each other [texture, colour, etc.]
🧠all theories have different viewpoints on whether or not the perspective in which the object is viewed
matters
○ view-dependent:orientationmattersin order to recognizean object
■ i.e.template theory
○ view-invariant:orientationdoes notmatterin orderto recognize an object
■ i.e.RBC theory
🧠repetition suppression (RS):brain activity decreaseswhen your brain gets used to looking at a
stimuli
○ repetition:repeated exposure
○ suppression:reduction in brain activity
○ this phenomenon was discovered shortly after the development of fMRI
○ to observe this phenomenon, you would use the subtraction method
■ [neural activity for 1st presentation] - [neural activity for 2nd presentation]
● present a stimulus for the first time, note the brain areas that activate [i.e.
attention, visual perception, etc.]
● take the stimulus away for a bit
● present the stimulus again and take note on the activated brain areas
○ you’ll notice that there would be less areas active since your brain is
getting used to the stimulus and has a better mental picture of it
○ this reduction in brain activitydoesn’t hinder recognitionability, rather is more about the
brain beingmore efficientwhen it recognizes objects
○ hypothesis: each time you see a stimulus the brainfine tunes the features that make up the
object, filters out unnecessary features [i.e. the colour of an umbrella not an important
defining feature of umbrella]
○ this theory can work for any visual stimulus [i.e. faces]
■ thebrain recalibrates when there’s a dramatic change[Kaaviya w/ straight blue hair]
🧠the fusiform area located in the temporal lobe is known for object recognition
🧠using RS to test out if perspective of an object’s orientation matters
○ fusiform area in the left hemisphere = view-invariant
○ fusiform area in the right hemisphere = view-dependent
○ ergo, the brain isboth view-invariant and view-dependent,depending on the side of brain
you’re looking at
