Reducing the problem of face recognition to an average February 5, 2008Posted by Johan in Applied, Cognition, Face Perception, Theory.
1 comment so far
Although computer software is now adept at face detection – Google’s image search does it, and so does you camera if you bought it within the past year – the problem of recognising a face as belonging to a specific individual has proved a hard nut to crack.
Essentially, this is a problem of classification. A model for this process should be able to sort images of three persons into three separate categories. This is remarkably difficult to do. If you look at the sheer physical differences between images of the same person, they easily outnumber the differences between images of different persons, taken from the same angle under the same lighting conditions. In other words, the bulk of the physical variability between different face images is uninformative, as far as face recognition is concerned. Thus, this remains an area where humans effortlessly outperform any of the currently-available face recognition models.
Recent work by Mark Burton at the Glasgow Face Recognition Group suggests a solution by which computer models can achieve human-like performance at face recognition. By implication, such a model may also offer a plausible mechanism for how humans perform this task. The model that Burton et al (2005) proposed is best explained by this figure, which outlines the necessary processing steps:
For each face that the model is to learn, a number of example images are collected (as shown in A). These images are morphed to a standard shape (B), which makes it possible to carry out pixel-by-pixel averaging to create a composite (C). This composite is then used by the model to attempt to recognise a new set of images of the person.
This may sound relatively straight-forward, but the idea is novel. Most face recognition models that work with photographs use an exemplar-based algorithm, where the model stores each of the images it is shown. Such models do improve as more faces are added (since there are more exemplars that might possibly match), but not as much as an averaging model does as more pictures are added to the average (Burton et al, 2005). Furthermore, when noise is added in the form of greater variations in lighting, the exemplar model breaks down rapidly while the averaging model is largely unaffected.
Why is this model so effective? The averaging process appears to remove variability that is not relevant to personal identity (such as differences in lighting and shading, changes in hair style), while preserving information that is informative for recognition (eyebrows, eyes, nose, mouth, perhaps skin texture). The figure at the top of this post highlights this (from Burton et al, 2005). The pictures are shape-free averages, created from 20 exemplar pictures of each celebrity. To the extent that hair is present, it is usually blurry. But the pictures are eminently recognisable, even though you have in fact never seen any of these particular images before (since they are composites). Indeed, Burton et al (2005) showed that participants were faster to recognise these averages than they were at recognising the individual exemplar pictures.
In the latest issue of Science, Jenkins and Burton (2008) presented an unusual demonstration of the capabilities of this model. They pitted their model against one of the dominant commercial face-recognition systems (FaceVACS). The commercial model has been implemented at MyHeritage, a website that matches pictures you submit to a database of celebrities.
Jenkins and Burton (2008) took advantage of this by feeding the website a number of images from the Burton lab’s own celebrity face database. Note that the website is all about matching your face to a celebrity, so if an image of Bill Clinton from the Burton database is given as input, you would expect the face recognition algorithm to find a strong resemblance to the Bill Clinton images stored by MyHeritage. Overall, performance was unimpressive – 20 different images of 25 male celebrities were used, and the commercial face algorithm matched only 54% of these images to the correct person. This highlights how computationally difficult face recognition is.
In order to see how averaging might affect the model’s performance, Jenkins and Burton (2008) took the same 20 images and created a shape-free average for each celebrity. Each average was then fed into the model.
This raised the hit rate from 54% to 100%.
The model that Burton is advocating is really one where individual face images are recognised with reference to a stored average. This finding is essentially the converse – the commercial model, which attempts to store information about each exemplar, is used to identify an average. But there is no reason why it wouldn’t work the other way around.
This demonstration suggests that as far as computer science is concerned, the problem of face recognition may be within our grasp. There are a few remaining kinks before we all have to pose for 20 passport pictures instead of one, however: the model only works if each exemplar is transformed, as shown in the figure above. As I understand it, this process cannot be automated at present.
While we’re on the computer science side I think it is also worth mentioning that there may be some ethical implications to automatic face recognition, especially in a country with one CCTV camera for every 5 inhabitants (according to Wikipedia). I have always dismissed the typical Big Brother concerns with the practical issue of how anyone would have time to actually watch the footage. If, however, automatic face recognition becomes common-place, you had better hope that your government remains (relatively) benevolent, because there will be no place to hide.
Turning to psychology, the assertion by Burton et al is that this model also represents to some extent what the human face recognition system is doing. This sounds good until you realise that face recognition is not hugely affected by changes in viewing position – you can recognise a face from straight on, in profile, or somewhere in between. This model can’t do that (hence the generation of a shape-free average), so if the human system works this way, it must either transform a profile image to a portrait image in order to compare it to a single, portrait average, or it must store a number of averages for different orientations, which leads to some bizarre predictions (for example, you should have an easier time recognising the guy who sits next to you in lecture from a profile image, because that’s how you have usually viewed him).
That being said, this model offers an extremely elegant account of how face recognition might occur – read the technical description of FaceVACS to get a taste for how intensely complex most conventional face recognition models are (and by implication, how complex the human face recognition system is thought to be). The Burton model has a few things left to explain, but it is eminently parsimonious compared to previous efforts.
Burton, A.M., Jenkins, R., Hancock, P.J.B., & White, D. (2005). Robust representations for face recognition: The power of averages. Cognitive Psychology, 51, 256-284.
Jenkins, R., Burton, A.M. (2008). 100% Accuracy in Automatic Face Recognition. Science, 319, 435. DOI: 10.1126/science.1149656
Can research be both relevant and fun? April 29, 2007Posted by Johan in Cognition, Economics.
add a comment
While most science bloggers were up in arms over Shelley’s successful campaign against Wiley, a bit of controversy has been stirring up over in economics. (I had no idea I was interested in economics, but judging by the amount of blogging that I’ve done on it, I am. Go figure) Noam Scheiber wrote an article in New Republic, subtly titled How Freakonomics is Ruining the Dismal Science. The article has now found its way online, thanks to a blogger who almost certainly is in violation of fair use, unlike the Retrospectacle head honcho.
For those of you who have somehow missed it, Freakonomics is a book that rogue economist Steve Levitt co-wrote with Steve Dubner. Essentially, it’s a collection of pop-science write-ups of studies Levitt has published over the years. This research, concerning unusual topics like the economics of drug-dealing and regression analyses that investigate whether sumo wrestling is rigged, turned out to have quite a bit of mass appeal, as Freakonomics promptly sold in marginally fewer copies than the Bible back in 2005.
Not everyone is so impressed. As the title hints at, Scheiber’s article is a scathing attack on Levitt’s research, with some borderline ad hominem elements. The article’s central thesis is that Levitt’s popular and academic success is part of a larger movement in economics that has had a dangerous influence on impressionable economics grad students. Apparently, they have now abandoned the rigorous and perhaps dull study of the macro-economy in favour of fast and fun studies of unusual topics, Freak-style. Scheiber argues that the consequence of this development is that method has become more important than theory. The studies no longer reveal anything of theoretical significance – instead they are novelties, getting attention not because of what they reveal, but because of how they reveal it. Oh, and along the way we also get to learn that Levitt has a squeaky voice and is a poor lecturer, in a perhaps less well-considered comment towards the end of the article.
Anyone who achieves success on the level of Levitt is bound to have a few scathing critics on the web, but the interesting bit about this particular case is that Levitt has responded to the Scheiber’s criticisms on the Freakonomics blog. Apart from responding to Scheiber’s ad hominems and pointing out a few inaccuracies (apparently, Scheiber does not have a PhD in economics and has never met Levitt, contrary to what his article seems to suggest), Levitt argues rather forcefully that the use of “clever” methods in no way precludes theoretical relevance. He points to a number of hard, real-life issues that his research has tackled (not citing the sumo study, surprisingly), in support of this claim.
In a way, Levitt is absolutely right. Many of the studies in Freakonomics are ones that, to quote the awarding criterion for the IgNobel prize that Levitt is sure to win sooner or later, makes you laugh and then think. For instance, a chapter in the book is dedicated to Levitt’s somewhat controversial notion that the vast drop in violent crime that the US experienced in the 1980′s and 90′s is a direct consequence of Roe vs. Wade, 10-20 years earlier. Levitt conjures up a range of statistics and deductive reasoning to support an argument that goes something like this:
1. If aborted fetuses are unwanted, the babies that were born before 1973 rather than being aborted were unwanted.
2. Unwanted children are at risk for crime and anti-social behaviour.
3. Thus, Roe vs. Wade meant that unwanted children were no longer being born at the same rate following 1973. This results in a drop in crime some 15-20 years later because that’s when the unwanted children would have otherwise started their criminal careers.
The argument is simple enough, but it is also quite original. Most people do have an initial visceral reaction to the notion of somehow equating unborn babies with potential criminals, but once you get past that point the idea is not entirely easy to refute.
To be fair, Scheiber has a point in that Levitt’s research is light on theory – this is something that Levitt himself admits to in Freakonomics. The controversial crime drop theory aside, most of the research in Freakonomics makes a practical point about real life, but cannot be easily fitted into the theoretical framework of economics. A lot of it is really best classified as sociology or political science. Perhaps part of the reason why Levitt seems to bother some economists is that he does this research as a professor of economics, often publishing his results in economics journals.
It’s not dissimilar to the way most empirically-based psychologists react to psychoanalysts, reflexologists, or even Dr Phil – by calling themselves psychologists, they contribute to a definition or a stereotype of psychology that many people in research detest. Much of the ire that both Levitt and Dr Phil receive from their peers is probably caused by the way they “make us look bad.” Neither one would get nearly as much of a reaction if they didn’t insist on calling themselves economist and psychologist, respectively.
Anyway, I wonder if they will be a Freakonomics of psychology. The best-selling psychology researchers, people like Pinker or Damasio, are perhaps better known for their style of writing and insight rather than for the sheer originality or wow-factor of their research. Still, there is some psychology research out there that would fit the bill – for one, Godden and Baddeley’s (1975) study on context dependency of memory comes to mind. In this study, divers encoded and recalled lists either over or under water, which produced a nice crosswise interaction: recall was superior when the encoding and recall context was identical, as can be seen below.
(Apologies for the poor quality)
Another prime example would be the (now numerous) studies that use a person in a gorilla suit to probe inattentional blindness (one example). The idea is to have the participants perform a demanding visual task, while casually letting a gorilla walk by. It is strikingly unusual that the participant reports having even seen the gorilla, when asked afterwards.
However, there is no real Levitt in Psychology, yet. Psychologists win IgNobels all the time, but it’s possible that most are too concerned with their reputation to don the gorilla suit for more than the odd study…
Godden, D., & Baddeley, A.D. (1975). Context-Dependent Memory in Two Natural Environments: On Land and Under Water. British Journal of Psychology, 71, 99-104.
Encephalon #20 at Neurontic April 9, 2007Posted by Johan in Cognition, Neuroscience.
add a comment
The new Encephalon is out, with a very nice write-up.
I particularly liked Madam Fathom’s account of a recent neuronal theory of sensory integration, using dodgeball as an example.
The Object Recognition Demons March 8, 2007Posted by Johan in Cognition, Off Topic, Sensation and Perception.
add a comment
This figure popped up in one of my lecture, and I thought it was a rather nice summary of how most theories of object recognition imagine that the process works (click for bigger version):
A bit easier to remember than your average boxes-and-arrows model, isn’t it! The traditional, functional box-and-arrow model is already being replaced by connectionist models, of course.. But I promise, once that fad blows over, the next paradigm is going to be little demons in your head. Psychologists will be scratching their heads, sweating over animation courses instead of programming languages.