[tokenadult]

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...

[bane]

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.

[bsenftner]

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).

[anigbrowl]

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.

[jisaacso]

Check out https://www.facebase.org/

Facial prediction is still in its infancy but is an exciting area of research.