Genetic mugshot recreates faces from nothing but DNA(newscientist.com)
newscientist.com
Genetic mugshot recreates faces from nothing but DNA
http://www.newscientist.com/article/mg22129613.600-genetic-mugshot-recreates-faces-from-nothing-but-dna.html
25 comments
Similar: Heather Dewey-Hagborg's project http://deweyhagborg.com/strangervisions/
In Stranger Visions artist Heather Dewey-Hagborg creates portrait sculptures from analyses of genetic material collected in public places. Working with the traces strangers unwittingly leave behind, Dewey-Hagborg calls attention to the impulse toward genetic determinism and the potential for a culture of genetic surveillance.
In Stranger Visions artist Heather Dewey-Hagborg creates portrait sculptures from analyses of genetic material collected in public places. Working with the traces strangers unwittingly leave behind, Dewey-Hagborg calls attention to the impulse toward genetic determinism and the potential for a culture of genetic surveillance.
What does "the impulse towards genetic determinism" mean here? Comparisons of identical twins raised apart show that physical appearance is highly heritable, so it would seem that "the impulse towards genetic determinism" is in this case correct.
See also:
"Heather talks about how a fixation with a single hair led to a controversial art project, the study of genetics, and the bones of an unidentified woman"
https://vimeo.com/71657839 http://deweyhagborg.com/strangervisions/portraits.html
"Heather talks about how a fixation with a single hair led to a controversial art project, the study of genetics, and the bones of an unidentified woman"
https://vimeo.com/71657839 http://deweyhagborg.com/strangervisions/portraits.html
"Yes, the perp is some Puerto Rican guy of average Puerto Rican height"
I want this in my next MMO character design system.
This is more like a facial composite[1] than a mug shot[2]. A mug shot is a photo of an identified person. A facial composite is a synthetic graphic of an unidentified person. The distinction is important because a facial composite is an investigative tool use to identify a suspect, and a mug shot is not.
[1] https://en.wikipedia.org/wiki/Facial_composite [2] https://en.wikipedia.org/wiki/Mug_shot
[1] https://en.wikipedia.org/wiki/Facial_composite [2] https://en.wikipedia.org/wiki/Mug_shot
It says right there that the purpose of a mug shot is also to identify a suspect. Even when the police gets a hold of someone and can take a photo there is no reason to think that identification is no longer necessary. Identification is probably what most criminal cases are about and being able to get to know someone’s name and to make a photo of them is not the actual relevant part of the identification for solving a crime.
Conceptually I would agree that this is more similar to facial composites (it’s a recreation of the suspect’s face not directly via a photograph but via some other means, either memory or, in this case, DNA) but comparing it to mug shots is not absurd or even weird. There is even a good argument one could make that this is conceptually more similar to mug shots than facial composites: both photos and DNA don’t rely on someone else via memory but directly on the suspect.
Since mug shots are well known making the comparison for the sake of a headline makes perfect sense. The term facial composite is much less well known. The conceptual similarity is there either way.
I think the comparison very clearly communicates why this new way of creating mug shots is very useful and as such there is nothing wrong with it.
Conceptually I would agree that this is more similar to facial composites (it’s a recreation of the suspect’s face not directly via a photograph but via some other means, either memory or, in this case, DNA) but comparing it to mug shots is not absurd or even weird. There is even a good argument one could make that this is conceptually more similar to mug shots than facial composites: both photos and DNA don’t rely on someone else via memory but directly on the suspect.
Since mug shots are well known making the comparison for the sake of a headline makes perfect sense. The term facial composite is much less well known. The conceptual similarity is there either way.
I think the comparison very clearly communicates why this new way of creating mug shots is very useful and as such there is nothing wrong with it.
New Scientist is a British magazine, but 'mug shot' isn't an everyday term in the UK (partly because UK police don't automatically release photos of arrestees like most US police), so over there the term is used a bit more loosely.
It's eerie how accurate they recreated her nose. It'll be interesting to see how soon this takes off
We are getting close to the yearly annoyance that is April fools' day, but this seems legitimate (a day would be fine, but it's starting to be more like a month, so I had to check that the New Scientists is published weekly). The article somewhat exaggerates the performance, though, by showing the copy-pasted hair, ears and earring (in gray). The PLOS paper (http://www.plosgenetics.org/article/info:doi/10.1371/journal...) doesn't show any of those. Ears in particular, seem a logical next target.
"One day, the technique may even allow us to gaze into the faces of extinct human-like species that interbred with our own ancestors."
I don't know how accurate that statement is. They can predict human faces because they can look at examples of DNA and the paired human face, they would have to use unsupervised machine learning for extinct species and have no real way to validate it.
I don't know how accurate that statement is. They can predict human faces because they can look at examples of DNA and the paired human face, they would have to use unsupervised machine learning for extinct species and have no real way to validate it.
This is possible as the molecular machinery for translating a DNA sequence into a 3-D structure is highly conserved - basically if you can convert modern human DNA into a face then you can convert other close relatives as well.
If you developed it to give consistently good results on other primates and then other mammals, then you could probably feel pretty confident about things such as cranial shape, eye color etc. etc.
We could probably feel pretty confident about eye color right now. What color are gorilla eyes? Howler monkey eyes? Humans barely vary in eye color at all; fancy European eye colors are a very recent phenomenon.
It's sort of like our earlier generations' expectations of privacy were set only by security-through-obscurity.
It would be helpful to see more validation examples like the photograph of the former New Scientist reporter shown with the article. I wonder how we would judge the resemblance if the photographs hadn't shown the additional gray outline (complete with earring in the same design!) with the computer-generated images.
I know plenty of examples of people from the same family lineage (siblings or first cousins) who grew up in different countries, and there is considerable influence of diet and other factors of childhood environment on people's appearance. For the computer-generated images, presumably the image-generating software is choosing a central-tendency value for the facial features predicted by the genetic samples, but for forensic purposes it would be important to know the "reaction range" for each gene assembly, as that reaction range may be quite large. For example, my two American nieces who are monozygotic twins were brought up in the same household by the same parents, but they do not look indistinguishably "identical," but rather can be told apart readily by their parents and other close relatives and told apart with careful thought by other people who know them. Genes have never been the whole story about how people look.
German monozygotic twins Otto and Ewald, who pursued two different sports and ended up with very different physiques,[1] are a classic example in genetics classes of how genes are not completely destiny for personal appearance.
[1] https://www.google.com/search?q=Otto+Ewald
http://thesameffect.com/check-out-identical-twins-otto-and-e...
I know plenty of examples of people from the same family lineage (siblings or first cousins) who grew up in different countries, and there is considerable influence of diet and other factors of childhood environment on people's appearance. For the computer-generated images, presumably the image-generating software is choosing a central-tendency value for the facial features predicted by the genetic samples, but for forensic purposes it would be important to know the "reaction range" for each gene assembly, as that reaction range may be quite large. For example, my two American nieces who are monozygotic twins were brought up in the same household by the same parents, but they do not look indistinguishably "identical," but rather can be told apart readily by their parents and other close relatives and told apart with careful thought by other people who know them. Genes have never been the whole story about how people look.
German monozygotic twins Otto and Ewald, who pursued two different sports and ended up with very different physiques,[1] are a classic example in genetics classes of how genes are not completely destiny for personal appearance.
[1] https://www.google.com/search?q=Otto+Ewald
http://thesameffect.com/check-out-identical-twins-otto-and-e...
I agree that diet and other environmental factors definitely contribute to appearance. For example, even though many Americans are largely European stock, there are certain developmental tendencies in some areas that I feel like it's possible to, with a better than chance guess and based on a little observation, tell if somebody is American or European - it can be something in the food, or differences in popular sports, or social body language, but all leading to developmental differences.
Of course I'm generalizing a bit, but I'm almost as good as my native South Korean wife at guessing where a person from East Asia is from. It's a combination of factors, height, facial features, but often as not it's fashion (which includes hair, eyebrows, facial hair, glasses style) etc.
An interesting test is found here http://alllooksame.com/
I can guess it pretty above average.
But it's a pretty big jump from guessing what stock a person is from to predicting what they look like based on their stock.
The problem with the examples in the article of course is that the example really looks very little like the actual woman. In fact, it's a pretty terrible likeness, especially in the eyes. About the only part that's debatable is the nose. I wouldn't rely on it for a police sketch.
Another challenge will be in populations with large groups of "mixed" children. Americans are a reasonable example, even though we're largely Europeans at this point in history, that's changing quickly, and even in the Caucasian population, there are very few who don't have ancestors from all over the place. What about my children? Will they have red hair or dark hair, epicanthic folds or not? Will they be barrel chested like my father's side, or have a physique more like my wife's? Even dominant genes can be suppressed in the right hormonal and developmental environment.
This sounds very sci-fi, but it's as long away off as those novelty "what will your children look like" facial morphers that were so popular in the 90s.
Of course I'm generalizing a bit, but I'm almost as good as my native South Korean wife at guessing where a person from East Asia is from. It's a combination of factors, height, facial features, but often as not it's fashion (which includes hair, eyebrows, facial hair, glasses style) etc.
An interesting test is found here http://alllooksame.com/
I can guess it pretty above average.
But it's a pretty big jump from guessing what stock a person is from to predicting what they look like based on their stock.
The problem with the examples in the article of course is that the example really looks very little like the actual woman. In fact, it's a pretty terrible likeness, especially in the eyes. About the only part that's debatable is the nose. I wouldn't rely on it for a police sketch.
Another challenge will be in populations with large groups of "mixed" children. Americans are a reasonable example, even though we're largely Europeans at this point in history, that's changing quickly, and even in the Caucasian population, there are very few who don't have ancestors from all over the place. What about my children? Will they have red hair or dark hair, epicanthic folds or not? Will they be barrel chested like my father's side, or have a physique more like my wife's? Even dominant genes can be suppressed in the right hormonal and developmental environment.
This sounds very sci-fi, but it's as long away off as those novelty "what will your children look like" facial morphers that were so popular in the 90s.
My work primarily based around 3D reconstruction of human faces from photos. It's a machine learning problem, and we have quite good results at www.3D-Avatar-Store.com. I have personally performed thousands of 3D reconstructions of people, and have developed a pretty good intuitive sense about a person from their facial photo.
Of particular interest has been reconstructing twins and near age siblings at different points in their lives. From my experience, the strongest drivers of facial appearance I've found are, in order:
1) genetics & ethnic background 2) attitude (not life style, but life perspective) 3) life style
In #1, I see very little variation between children of similar ethnic backgrounds. And when I say "little variation", I mean in the formal sense, from the facial recognition coefficients. Children are interesting in their facial photos often clearly indicate how they will progress through #2.
I'm yet to read formal observations of my #2; It is very clear to me when creating reconstructions of real people, it is their "attitude" they wear on their face day after day that by the time an individual reaches adulthood that "look" is impressed on their face. And this "look" tends to increase over time and age, unless a major attitude adjustment takes place.
I see #3 as a strong influence to facial appearance, but it is driven by #2, so I consider attitude to be the #1 driver of facial appearance which an individual can have control over (short of surgery).
Of particular interest has been reconstructing twins and near age siblings at different points in their lives. From my experience, the strongest drivers of facial appearance I've found are, in order:
1) genetics & ethnic background 2) attitude (not life style, but life perspective) 3) life style
In #1, I see very little variation between children of similar ethnic backgrounds. And when I say "little variation", I mean in the formal sense, from the facial recognition coefficients. Children are interesting in their facial photos often clearly indicate how they will progress through #2.
I'm yet to read formal observations of my #2; It is very clear to me when creating reconstructions of real people, it is their "attitude" they wear on their face day after day that by the time an individual reaches adulthood that "look" is impressed on their face. And this "look" tends to increase over time and age, unless a major attitude adjustment takes place.
I see #3 as a strong influence to facial appearance, but it is driven by #2, so I consider attitude to be the #1 driver of facial appearance which an individual can have control over (short of surgery).
Physique can obviously vary widely, and differences in nutrition at key stages of development can have noticeable effects on facial bone structure and dentition, but I suspect the basics of facial geometry won't drift too far in most cases. It would be very interesting to take the generated output from this and see how it performs against some commercial facial recognition algorithms that has been trained on a large dataset, eg Facebook's.
Check out https://www.facebase.org/
Facial prediction is still in its infancy but is an exciting area of research.
Facial prediction is still in its infancy but is an exciting area of research.
Quite interesting to play with to get an interactive feel for how the various parameters of their model affect the facial appearance. You really need to read the paper to figure out what's going on though.
[1] http://www.plosgenetics.org/article/info:doi/10.1371/journal...
[2] https://dl.dropboxusercontent.com/u/12163246/PC2014/DNA2FACE...