| View previous topic :: View next topic |
| Author |
Message |
baylor
Guest
|
 Posted: Wed Jul 16, 2003 3:08 am Post subject: Q: on feature extraction & object recognition |
 |
|
i hope this is the right newsgroup for this post. It>s not a flame.
Hope that>s alright :)
i>m looking for a pointer to books, papers, Web sites, software,
search terms for Google, etc. on how people distinguish between
objects (situations, movements, etc.)
While driving down the road pointing out cars to a friend ("that>s an
'87 Sentra, that must be the new Subaru Impreza, that>s a Honda Civic
with aftermarket lights and exhaust, i>m not sure that is but it looks
like a Volkswagen"), i wondered why it was that i, like most of my
friends, could look at most any car and determine the manufacturer,
model and often the year to within 2 years. After all, cars look
mostly alike. Same number of wheels, tires, windows, sideview mirrors,
etc.
i thought maybe it was like a decision tree where information gain was
used to find the most important features and distinguish objects by
those. But then i realized that i was determing which cars were which
by looking at the back of the cars - rear bumper, brake lights, turn
lights and trunk lid. A normal AI algorithm like DT, ANN, SVM or CBR
normally asks that you specify all attributes of an object before it
classifies it. So it must not be like the typical classification AI
task
Then i remembered reading in a Pinker book that some people people
believe we store snapshots of an object in our mind. Specifically, we
store several images, one from each significant angle (like 25 i
think). My CogSci book talked about reaction time studies and
canonical forms and said the closer the angle was to how we typically
see it the faster we recognize the object. In their example, people
recognized a profile shot of a horse faster than the recognized an
above shot. But the point is that maybe i had stored images of car
rear views and matched that image to the concept of that specific car
It caused me to think of more things like chunking and 7+/- 2 and
proceduralization and how people supposedly can>t read (comprehend)
sentences until they>ve memorized and proceduralized the ability to
read letters and then words. When i saw a car, maybe processing
happened at different levels. At one level is a bunch of lines (a tail
light has 4 lines). At another level is a collection of shapes (tail
lights are 2 rectangles). Is the next level "car"? Or are there more
levels like this square plus this square and those two squares over
there are the tail lights, that L-shaped curved section of metal is
the trunk lid, etc. and a car rear end is made of lights, a bumper and
a trunk and a car is made of a front, sides, top, windows and a tail.
And if we do analyze shapes as a hierarchy of feature groups how do we
decide to group things? By physical proximity? Lines/edge tracing? By
knowledge of purpose/functional grouping?
When i see the car, i instantly say "toyota, ford, chevy". i don>t
remember any of the details. i don>t conciously remember seeing tail
lights or tires or a sloped front windshield. i maybe did back when i
was learning about cars, but not any more. Can recognition of objects
i once paid a lot of attention to be proceduralized?
Anyway, i imagine lots of peole have studied how humans recognize
objects (or scenes or themes or movement types). But it>s not in my
cogsci book or any of my AI stuff and i don>t know the proper
terminology to search for it on Google.
Help?
-baylor |
|
| |
|
 Back to top |
Ken Catchpole
Guest
|
| Posted: Wed Jul 16, 2003 1:19 pm Post subject: Re: on feature extraction & object recognition |
|
|
"baylor" <baylor@no_spam.ihatebaylor.com> wrote in message
news:k6t8hvk7rns6vqj9h728ekdcv69a9auc1o@4ax.com...
[quote]i hope this is the right newsgroup for this post. It>s not a flame.
Hope that>s alright :)
i>m looking for a pointer to books, papers, Web sites, software,
search terms for Google, etc. on how people distinguish between
objects (situations, movements, etc.)
[/quote]
<Lots of very good stuff snipped>
Nice to have an intelligent thread on ths NG for once.
It sounds like you>ve already got a fair idea of what sorts of things are
involved. AI will offer a perspective, but IMHO it is unlikely to offer a
great deal of insight into human object recognition as humans and machines
behave in very different ways. Unfortunately, I don>t have any of my psych
reference books to hand, but you might want to try:
Principles of cognitive psychology / Michael W. Eysenck. - 2nd
ed.Principles of cognitive psychology / Michael W. Eysenck. - 2nd ed
(2001).
High-level vision : object recognition and visual cognition / Shimon
UllmanHigh-level vision : object recognition and visual cognition / Shimon
Ullman (1996). |
|
| |
|
| Back to top |
Glen M. Sizemore
Guest
|
| Posted: Thu Jul 17, 2003 4:11 pm Post subject: Re: on feature extraction & object recognition |
|
|
...
Say you wanted produce an animal (say a pigeon) that
could select photos containing trees. How would you
do it. How would you show that you had not simply
produced a bunch of separate response classes? This is
where the study of "categorization" should start.
"Ken Catchpole" <k.catchpole@ich.ucl.ac.uk> wrote in message news:<bf31n9$bj6$1@uns-a.ucl.ac.uk>...
[quote]"baylor" <baylor@no_spam.ihatebaylor.com> wrote in message
news:k6t8hvk7rns6vqj9h728ekdcv69a9auc1o@4ax.com...
i hope this is the right newsgroup for this post. It>s not a flame.
Hope that>s alright :)
i>m looking for a pointer to books, papers, Web sites, software,
search terms for Google, etc. on how people distinguish between
objects (situations, movements, etc.)
Lots of very good stuff snipped
Nice to have an intelligent thread on ths NG for once.
It sounds like you>ve already got a fair idea of what sorts of things are
involved. AI will offer a perspective, but IMHO it is unlikely to offer a
great deal of insight into human object recognition as humans and machines
behave in very different ways. Unfortunately, I don>t have any of my psych
reference books to hand, but you might want to try:
Principles of cognitive psychology / Michael W. Eysenck. - 2nd
ed.Principles of cognitive psychology / Michael W. Eysenck. - 2nd ed
(2001).
High-level vision : object recognition and visual cognition / Shimon
UllmanHigh-level vision : object recognition and visual cognition / Shimon
Ullman (1996).[/quote] |
|
| |
|
| Back to top |
Isla Kul XII
Guest
|
| Posted: Fri Jul 18, 2003 2:27 pm Post subject: Re: on feature extraction & object recognition |
|
|
baylor <baylor@no_spam.ihatebaylor.com> a écrit dans le message :
k6t8hvk7rns6vqj9h728ekdcv69a9auc1o@4ax.com...
[quote]i hope this is the right newsgroup for this post. It>s not a flame.
Hope that>s alright :)
i>m looking for a pointer to books, papers, Web sites, software,
search terms for Google, etc. on how people distinguish between
objects (situations, movements, etc.)
[/quote]
[snip]
[quote]Anyway, i imagine lots of peole have studied how humans recognize
objects (or scenes or themes or movement types). But it>s not in my
cogsci book or any of my AI stuff and i don>t know the proper
terminology to search for it on Google.
[/quote]
Ask for "expertise in object and face recognition" in Google. This is the
title of a paper from James Tanaka and Isabel Gauthier. You should find it
quite easily in PDF format.
It seems that faces and cars are identified in much the same way, and
roughly in the same place in our brains. So you might be intersted in face
recognition processes.
I am interested in face recognition impairment (or "face blindness") in
humans.
It is said it is a rare neuropathologic condition that cannot be cured. But
I would be very intersted in any (documented) method to improve my ability.
I also have a lot of problems with cars !
I hope you>ll find what you are looking for !
Isla |
|
| |
|
| Back to top |
Glen M. Sizemore
Guest
|
| Posted: Sat Jul 19, 2003 10:17 pm Post subject: Re: on feature extraction & object recognition |
|
|
> ...
[quote]
Say you wanted produce an animal (say a pigeon) that
could select photos containing trees. How would you
do it. How would you show that you had not simply
produced a bunch of separate response classes? This
is
where the study of "categorization" should start.
[/quote]
KC: I read some of your critique of cognitive research
in the earlier thread.
While I appreciate the difficulty of inference, your
arguments regarding the
"uselessness" of cognition is unconvincing.
GS: I>m not picking a fight here, Ken, but I hope I
didn>t say there was a problem with inference. The
problem concerns what is inferred. Skinner>s position
was that, under many circumstances, human behavior
(particularly human, in this case) cannot be explained
without inferring some sort of private behavior (i.e.,
behavior that is observable by the behaving individual
alone). So-called "cognition" is closely associated with
"what we are aware of," but the "cognitive processes"
responsible for "what we are aware of" are not,
themselves, observable to the behaving individual. It is
these alleged processes to which I am objecting. I
won>t elaborate yet (but see below).
KC: Though my lack of familiarity
with behaviourism prevents me from constructing a
counter claim, I can offer
two sensible perspectives.
1. There is a risk of arguing over definintion. I imagine
behaviourists and
congnitivists refer to similar phenomena - they just call
them different
things.
GS: It should be clear now that that isn>t true.
KC: Something underlies behaviour. The study of
cognition is a
structured attempt to understand whatever those things
are, which may or may
not help in identifying the behavioural affects of
physilogical processes in
the brain.
GS: I agree......but with the emphasis on "...may not
help in identifying the behavioural affects of physilogical
processes...," though I would not say it that way. In any
event, physiology underlies behavior, but I assert that the
fanciful, metaphorical mechanisms of cognitive "science" have
done nothing but impede the study of brain/behavior relations.
KC: 2. At the very least, cognition has served as a way
in which to understand
the world that has (for example) led to better design
and increased saftey
in human-machine systems.
GS: You mean because "ergonomics" occasionally
makes reference to cognitive nonsense?
KC: 3. It>s only sensible that the cognitivism and
behaviourism two co-exist.
Clearly, your work with animals is largely behaviourist.
But if you choose
to ignore cognition entirely (which seems to be the
position that you wish
to argue yourself into),[...]
GS: I don>t deny the behavioral phenomena SAID TO
REQUIRE "COGNITIONS," (and that includes "the
subjective") but I argue that "cognitions" are simply
explanatory fictions for which there exist no
independent observability. That is, "cognitions" are
inferred EXCLUSIVELY from observations and then
said to explain the observations from which they are
inferred.
KC: [...]you are throwing away an enormously valuable
resouce to you as a researcher. All research can be
learned from.
GS: I don>t think that is true, though that is the standard
psychological line. The problem is that worthless
concepts fail to constrain psychological theories. NOT
ALL QUESTIONS ARE EMPIRICAL. Most of
mainstream psychology is composed of concepts that
can>t stand serious scrutiny. But most psychologists
have not been taught to examine their assumptions, or
trained in the pitfalls of metaphors that can never be
anything BUT metaphorical.
KC: To answer your question, it depends on the
hypothesis that I wished to
construct.
GS: That doesn>t answer my question. The answer is
well known and has nothing to do with an hypothesis.
"Ken Catchpole" <k.catchpole@ich.ucl.ac.uk> wrote in message |
|
| |
|
| Back to top |
Joseph Legris
Guest
|
| Posted: Sun Jul 20, 2003 5:26 am Post subject: Re: on feature extraction & object recognition |
|
|
Glen M. Sizemore wrote:
....
[quote]In any
event, physiology underlies behavior, but I assert that the
fanciful, metaphorical mechanisms of cognitive "science" have
done nothing but impede the study of brain/behavior relations.
[/quote]
The information processing view, which is the main thrust of cognitive
science, has been successfully applied to complex behaviour in machines
and natural phenomena. It is only logical to apply it to the behaviour
of humans and animals. Your faith-based objections, which are as easy to
conjure as any fanciful metaphor, are both irrefutable and irrelevant.
--
Joe Legris |
|
| |
|
| Back to top |
Glen M. Sizemore
Guest
|
| Posted: Sun Jul 20, 2003 4:50 pm Post subject: Re: on feature extraction & object recognition |
|
|
...
JL: The information processing view, which is the main
thrust of cognitive
science, has been successfully applied to complex
behaviour in machines
and natural phenomena.
GS: Cognitive science has turned "information
processing" into the equivalent of "when this neuron
fires, the neuron it is connected to might fire." It is true
that this is an important notion, but it has more to do
with Sherrington than Shannon. The connection
between cognitive "science" and Shannon is simply that
by hand-waving one give the impression that one is
connected to a powerful mathematical apparatus, and
that telegraphy (minus the human sender and receiver)
is the same thing as the behavior of animals.
Cognitive science has simply re-bought the input-output
of S-R psychology and wrapped it up in the computer
metaphor.
JL: It is only logical to apply it to the behaviour
of humans and animals.
GS: Logic has little to do with it. The processes that
produce metaphorical extension are far more relevant.
Any logical "justification" is after the fact, and laden
with dubious premises.
JL: Your faith-based objections, which are as easy to
conjure as any fanciful metaphor, are both irrefutable
and irrelevant.
GS: Of course, you will not, or cannot, refute my
assertion that both behaviorism and cognitive "science"
rest on assumptions. One difference between them is
that behaviorism is largely an effort to illuminate
assumptions and adopt the most parsimonious set.
Another difference is that every time a cognitive
"scientist" does an experiment, they assert that they are
testing their assumptions; behaviorists know what is
assumption, and what is hypothesis.
Most cordially,
Glen
Joseph Legris <jalegris@xympatico.ca> wrote in message |
|
| |
|
| Back to top |
Joseph Legris
Guest
|
| Posted: Sun Jul 20, 2003 10:33 pm Post subject: Re: on feature extraction & object recognition |
|
|
Glen M. Sizemore wrote:
[quote]...
JL: The information processing view, which is the main
thrust of cognitive
science, has been successfully applied to complex
behaviour in machines
and natural phenomena.
GS: Cognitive science has turned "information
processing" into the equivalent of "when this neuron
fires, the neuron it is connected to might fire." It is true
that this is an important notion, but it has more to do
with Sherrington than Shannon. The connection
between cognitive "science" and Shannon is simply that
by hand-waving one give the impression that one is
connected to a powerful mathematical apparatus, and
that telegraphy (minus the human sender and receiver)
is the same thing as the behavior of animals.
Cognitive science has simply re-bought the input-output
of S-R psychology and wrapped it up in the computer
metaphor.
JL: It is only logical to apply it to the behaviour
of humans and animals.
GS: Logic has little to do with it. The processes that
produce metaphorical extension are far more relevant.
Any logical "justification" is after the fact, and laden
with dubious premises.
JL: Your faith-based objections, which are as easy to
conjure as any fanciful metaphor, are both irrefutable
and irrelevant.
GS: Of course, you will not, or cannot, refute my
assertion that both behaviorism and cognitive "science"
rest on assumptions. One difference between them is
that behaviorism is largely an effort to illuminate
assumptions and adopt the most parsimonious set.
Another difference is that every time a cognitive
"scientist" does an experiment, they assert that they are
testing their assumptions; behaviorists know what is
assumption, and what is hypothesis.
Most cordially,
Glen
[/quote]
You may be the originator of the idea that telegraphy (minus the human
sender and receiver) is the same thing as the behavior of animals. I
have never heard anyone else say that. How might it work?
The information processing view encourages the decomposition of a
complex system into a collection of interacting subsystems, each with a
set of inputs and outputs, but you have forgotten to consider what>s
inside each box. The telegraphic analogy is simply not up to the task
because it provides no mechanisms for data storage, transformation or
for adaptation over time. The power and success of the information
processing view comes from its inherent ability to provide these
characteristics in quantifiable and realizable ways using the results of
computer science and mathematics. In principle, every claim, abstract or
concrete, is backed up an implicit procedure for implementing the
behaviour on a machine. Try holding radical behaviourism to a similar
standard.
--
Joe Legris |
|
| |
|
| Back to top |
Eray Ozkural exa
Guest
|
| Posted: Wed Jul 30, 2003 11:43 pm Post subject: Re: on feature extraction & object recognition |
|
|
I>d like to point out that behaviorism is an obsolete doctrine and
Glen is a moron. Look at the way he posts on this group. He argues
with everybody in his uncivilized style, doesn>t contribute to any
theory, constantly repeating operant conditioning bullshit.
Glen might be just as smart as a pigeon in a Skinner box, but we>re
looking to solve problems related to intelligence, not mere behavior
from which we can infer a tiny fraction of a cognitive system.
Just because of idiots like Glen we should set up a global scientific
discussion network that is more advanced than USENET.
__
Eray Ozkural
Comp. Sci. Dept., Bilkent Univ.
gmsizemore2@yahoo.com (Glen M. Sizemore) wrote in message news:<6e2f1d09.0307210912.1d87ebf5@posting.google.com>...
[quote]GS: Yeah.......some standard. "Here>s a bunch of
made-up metaphorical shit that can>t be directly
observed sort of like phlogiston but you should believe
it anyway because we>re all a bunch of really clever
guys and I mean oh-so-much more clever than those
silly behaviorists and even though we are not directly
observing cognitive processes we can operationize
them by measuring behavior and simply claim that
we>re measuring cognitions." You mean that standard?[/quote] |
|
| |
|
| Back to top |
|
