Something else to think about

[Ed62]

For the most part, we all have 2 eyes, 2 ears, a nose, and a mouth. How many different looks can they have? Think of all the people you know or see. They all look different, except for identical twins. For the most part, we identify people by looking at their faces. OK Uncle C. I'm sure you'll have something to add to that.

Ed

[Uncle Cracker]

For the most part, we identify people by looking at their faces. OK Uncle C. I'm sure you'll have something to add to that.

Only that I'm glad we do it that way... Dogs gotta sniff butts...

[Pappy]

For the most part, we identify people by looking at their faces. Ed

There are exceptions to that rule. Dolly Parton comes to mind!

[Uncle Cracker]

There are exceptions to that rule. Dolly Parton comes to mind!

Dolly Parton has a face???

[woodturner]

For the most part, we all have 2 eyes, 2 ears, a nose, and a mouth. How many different looks can they have?

Less than you might expect. If we take 200 random human faces and decompose them through a simple mathematical process into 200 "components", then any other face can be represented by a 200 element vector (a list of 200 numbers) to a sufficient degree that a computer can identify an individual with better than 99% accuracy (people identify correctly about 98% of the time).

[cabinetman]

Dolly Parton has a face???

I would almost bet on it.
.

[Ed62]

Dolly Parton has a face???

I heard that she actually does! It's just that nobody's seen it.

Ed

[Uncle Cracker]

I heard that she actually does! It's just that nobody's seen it.

I'd be willing to bet there's no wrinkles in it...

[crokett]

Less than you might expect. If we take 200 random human faces and decompose them through a simple mathematical process into 200 "components", then any other face can be represented by a 200 element vector (a list of 200 numbers) to a sufficient degree that a computer can identify an individual with better than 99% accuracy (people identify correctly about 98% of the time).

It works because the given area for arranging the facial features is fairly limited. So while space between the eyes might vary from person to person, it isn't that big a difference between the lower and upper extremes.

[Mr__Bill]

For the most part, we identify people by looking at their faces.

There is also a great deal of in context identification too. Now if I were to see Dolly in church I would recognize her but the girl at the checkout in the grocery store when seen in church I may not recognize and would be filed as a new ah err, face. This might hold true for guys too but then, I don't pay much attention to guys.


Bill, thinking of summer at the beach.....

[woodturner]

It works because the given area for arranging the facial features is fairly limited. So while space between the eyes might vary from person to person, it isn't that big a difference between the lower and upper extremes.

It's not the relationship between the features but the features themselves that are used for identification.

Early research explored using the distance between features, for example, the distance between the eyes and the nose, to identify faces. It did not work well - when the tolerance of the measurement was increased to the point that a face that was turned slightly could be recognized, it was confused with other faces.

In other words, the variance between facial distance measurements for faces of different individuals is in the range or less than the variance for multiple images of the same individual.

The eigenface method I described is a wholistic method - it uses whatever information is present and different between the training faces and captures that differentiating information.

[scmhogg]

I'm glad computers are good at facial recognition, because we humans are not. The main cause of convictions of the innocent is eyewitness identification. With the advent of DNA, this has become more apparent.

Here, in California, a man was convicted and imprisoned, after being identified as the one who inflicted a severe beating and rape of his WIFE. DNA proved another committed the crime. The culprit looked nothing like her husband and was of a different race.

Steve

[gsmittle]

Less than you might expect. If we take 200 random human faces and decompose them through a simple mathematical process into 200 "components", then any other face can be represented by a 200 element vector (a list of 200 numbers) to a sufficient degree that a computer can identify an individual with better than 99% accuracy (people identify correctly about 98% of the time).

Huh?

g.

[dbhost]

Dolly Parton has a face???

Actually yes, and considering her age, a quite appealing one at that...

[cgallery]

Dolly Parton has a face???

She is two-faced. Two of the biggest faces you've ever seen, in fact.