Rob Jenkins Face recognition, says Jenkins, is more about face perception, that is, getting a computer to not only detect a face, but also determine a person's gender, age, emotion, gaze direction and a host of other variables that the human brain computes almost instantly.
Many Moods of Bill Clinton Not every photo of Bill Clinton represents his face exactly the same way. Seemingly small variations in lighting and expression can lead to big mistakes in computerized recognition.
Law of Averages Jenkins' face perception program combines many images of the same person, in this case Bill Clinton, into one average photo that a computer has a better chance at recognizing.
9:32 AM imtracynotstacy: hello there
cognition101: Hi Tracy
imtracynotstacy: is this time still good for you?
9:33 AM cognition101: Absolutely. Do I look out of touch if I use capital letters?!
imtracynotstacy: No!
cognition101: good answer
9:34 AM imtracynotstacy: Ha. So let's start out talking a little bit about you. Where are you? And what's your area of research?
9:35 AM cognition101: I'm at the Department of Psychology, University of Glasgow, in sunny Scotland. And it is in fact sunny.
And my area of research is mostly face perception
imtracynotstacy: like face recognition?
9:37 AM cognition101:
Yes, I use the term face perception more broadly. Face recognition
normally refers to identifying an individual, whereas face perception
could include detection of a face, determination of the person's
gender, age, emotion, gaze direction and a host of other things. We
read a lot of information from the face!
imtracynotstacy: and why study face perception? for what applications would be it useful?
9:40 AM cognition101:
Well, to be honest, the reason I study it is because it interests me,
rather than for applied purposes. I want to know how the mind works,
and this is an interesting corner of it for various reasons. Happily
though, there are a good number of applications. The one that has
traditionally been dominant is forensic face recognition in the context
of eye witness testimony. More recently, security applications have
also been emphasized.
9:41 AM imtracynotstacy: can you say more about how this research helps you understand how the mind works?
9:44 AM cognition101:
It's a part of the more general problem of how the brain organises the
information arriving at the senses into meaningful chunks. Because
faces are complex stimuli and play a major role in our social lives,
it's a rich seam to tap. Also, some people have real difficulty in
reading information from faces. Where our abilities go awry can often
reveal something about their normal function.
9:45 AM imtracynotstacy:
so you try to get computers to perceive faces the way the brain does
and in the process, you gain some insight into how our brain functions.
IS that right?
9:48 AM cognition101:
Trying to get computers to behave in the same way is one common
approach. Ideally, you end up in a virtuous circle where the computer
model makes some novel prediction about human performance, so then you
go and test that in humans and then refine the model accordingly, and
gradually you ratchet up your understanding of what's going on. Like I
said, "ideally"! In practice it can be more like a plate of spaghetti
than a virtuous circle!
9:49 AM imtracynotstacy: So what are the tools of your trade? What pieces of tech do you rely upon to do your research?
9:52 AM cognition101:
Actually, the hardware side of things is pretty simple - cameras and
computers for the most part. Video cameras or still cameras to capture
face images, and computers to process the images, build models, test
experimental volunteers, and all the rest of it. We're starting to
incorporate some neuroimaging studies into the psychology experiments,
and some electrical engineering into the automation, but the core of it
is really quite basic.
imtracynotstacy: Are you in your lab now? or office?
9:55 AM cognition101:
I'm in my office right now. It's been an admin intensive week!
Although, it's unusually hot and sunny today, so it's better to be here
in front of an open bay window than down in the dark windowless labs.
imtracynotstacy: Glad you have so much sun! So what's the weirdest thing in your office?
9:59 AM cognition101:
Wow, errmm... looking around, I would say it's a tie between my giant
hourglass that actually runs for an hour (which was a 21st birthday
present from my father) and a geometrically very weird Guido Moretti
sculpture (which was second prize in the Best Visual Illusion of the
Year contest, if you can believe there is such a thing).
Now I keep seeing things all around me in varying degrees of weirdness.
imtracynotstacy: ha ha
do you listen to music when you are working?
cognition101: Sorry, do you want me to turn it down?
imtracynotstacy: I can hear it all the way across the pond!
10:02 AM cognition101:
Yes, I listen to music a lot here. In fact, I recently got some fancy
computer speakers and what not to make the most of it. I have a 20
month old daughter at home, so the domestic listening hours have been
somewhat curtailed. Although her response to music is a joy to behold.
imtracynotstacy: so what do you listen to?
10:05 AM cognition101:
Most of it is what I've come to describe as "very assertive" - grind
and death metal, extrapolate punk 30 years forward, and there you have
it. Get through that admin in half the time, and beat the parental
insomnia too!
imtracynotstacy: Do you incorporate music into your research at all?
10:11 AM cognition101:
Not really. There is a psychology of music literature, but it has
always seemed to me to approach the topic from a very strange angle. I
sometimes use musical analogies to communicate points though. For
example, caucasians sometimes comment that all oriental faces look
alike. What they often don't appreciate is that all caucasian faces
look alike to oriental people. To me, this is very much like acquiring
expertise in musical genres. If you listen to jazz all day and know
nothing about opera, you'll know Coltrane from Mingus after the first
note, but all opera will sound the same.
10:12 AM imtracynotstacy: What does that say about how our brain perceives things? Whether images or sound?
10:16 AM cognition101:
Well, I think one simple but quite profound conclusion is that we need
expertise in a domain in order to make fine discriminations within that
domain. And the expertise comes as a natural consequence of being
exposed to the types of variation in that domain. This is absolutely
key to face recognition research as far as I'm concerned. We're
fantastically good at telling our friends and acquaintances apart, but
we're hopeless with unfamiliar faces. So the expertise seems to
accumulate on a face-by-face basis. We're not experts with faces in
general.
10:18 AM imtracynotstacy:
So we need some experience with a face. We need to see it a bunch of
times and maybe in different scenarios before we can really pick it out
of a crowd. Is that right? This gets to the bit of research I wrote
about on Discovery earlier this year, right?
10:24 AM cognition101:
That's exactly it. We somehow need to grasp all the different ways that
particular face can look - when smiling, when gloomy, when looking up
or around, under disco lights, in the sunshine, when thrown into
shadow, when captured on video or on a polaroid. The amount of
variation is huge, and the central problem is that it can swamp the
physical variability between different people's faces, which are all
much the same structurally. That is the obstacle that is blocking
progress in automatic face recognition. Since the human brain somehow
solves it quite quickly with no apparent effort, face learning in
humans seems like a good lead.
10:25 AM imtracynotstacy:
so somehow the brain is able to take all of the variability (john sad,
john happy, john tired, john angry, john confused, etc.) and meld it
into meaningful information (that's john!).
10:33 AM cognition101:
Yes, that's our take on it. And notice that the resulting
representation is not a photograph. The standard approach to automatic
face recognition involves matching the incoming image against a
database of stored photographs. But even the mighty human brain, that
can walk and talk and make jokes, can't match photographs of unfamiliar
faces. What hope do we have of getting machines to do it? (given our
success with walking talking joking robots!). So the thinking is - move
beyond photographs. Once you start thinking about it that way, it makes
a lot of sense. Photos of faces have only been around for a hundred
years, so although they are socially ubiquitous, they're really quite
unnatural things to expect the brain to manage.
10:34 AM imtracynotstacy: it sounds really great and totally fascinating. I'm about to start my 10:30 meeting, so have to wrap this up.
cognition101: Okay, well it's been a pleasure talking to you. It's my first IM chat actually.
imtracynotstacy: tell me what's on your plate, research-wise, for the next year. What will you be working on?
cognition101: Over
the next year, there will be some further refinement of the work we've
just been talking about. I'm also interested in how attention affects
our perception of time, so that will be a new angle for me.
10:39 AM imtracynotstacy:
speaking of attention...I've started my conf call. :) so I'll let you
go. Thanks so much for your time. And good luck with your research.
10:40 AM cognition101: Many thanks - it was a fun interview, and a great idea for an interview format. Bye for now.
imtracynotstacy: goodbye! |
advertisement
Download the Tech WidgetWhat's On Now
|
