MetaChat - Ask MeCha Redux:

MetaChat is an informal place for MeFites to touch base and post, discuss and chatter about topics that may not belong on MetaFilter. Questions? Check the FAQ. Please note: This is important.

28 March 2007

Ask MeCha Redux: What is the fewest number of pixels (height by width as well as colors in the pallette) required to create a photorealistic image of a human face, allowing for other elements in the image like setting, background and/or objects associated with that person? Specifically, the smallest picture that the average person would think real?

(This question has its origin in an existential debate so subjective answers are as valid as hard numbers. Thankee, sai.)

post by: mischief at: 21:06 | 15 comments

Also, this particular debate sprang from a discussion of quantum phenomena occurring within the human realm of experience, so information theory is also welcome but not necessary.

posted by mischief 28 March | 21:10

A study! Seems to be exactly what you're looking for, I think.

posted by Zack_Replica 28 March | 21:14

well, except for the colour bit..

posted by Zack_Replica 28 March | 21:16

Hmm, how about a Photo Friday where we try to reduce famous portraits to their smallest size (height * width * pallette) yet remain identifiable by over half of everybody.

posted by mischief 28 March | 21:46

Hmm, or should that be pallette ^ (height * width)...

Thanks, Zack!

posted by mischief 28 March | 21:47

ZR's response is about the minimum resolution that's recognizable as a face. I took your question to mean an image of decent (i.e., near-photographic) quality. It may interest you that flickr's avatars (sample here) are 48 pixels square.

posted by rob511 28 March | 21:48

or perhaps not!

posted by rob511 28 March | 21:49

np, mischief!

posted by Zack_Replica 28 March | 22:27

Are you asking for the smallest picture that the average person would think was a photo?

posted by GeckoDundee 28 March | 22:34

Apparently 13x13 pixels are enough... gah, can't find the link right now

posted by porpoise 28 March | 23:40

Gecko: Yeah, not just a photo but an untouched photo. Zack's link provides a good baseline from which to work.

posted by mischief 28 March | 23:48

Favicons are 16x16, right?

These sites all have "photos" as favicons.

So 13x13? Sure, why not?

On preview, so somewhere between 48x48 and 16x16. Hmm... Interesting.

posted by GeckoDundee 28 March | 23:51

The "Automatic" and "Yasmeen" favicons are better than the "Joel" and "Adactio" ones - because they're more zoomed in and don't have a background. More space for the face, hence more information.

posted by matildaben 28 March | 23:58

In cog sci, a common size seems to be 40x40, but if you allow partial faces (which I doubt you would, given what you said about contextual elements) then 20x20 should work too. I know that "what an aNN can identify as a face" isn't the same as what "the average person would think real", but have a look at the samples. (Skip the second one though, it's a huge file.)

On preview again, what mats says goes to the requirement you make about it having context around it too.

posted by GeckoDundee 29 March | 00:03

I've seen 14x14 greyscale as a recognizable face. However we all guess wrong, so I'm thinking that we are close to a borderline there

posted by eriko 29 March | 06:53

← Ask MeCha: || Taxing work →

HOME || REGISTER || LOGIN