Statistics of the colors in natural images

Re: Statistics of the colors in natural images?

On Dec 5, 5:57=A0am, AE lover  wrote:
> Hi guys,
>
> Do you know if there is any paper discussing statistics of colors in
> natural images?
>
> For example, if there is anyone trying to analyze a database of
> natural images and do some statistics to get information about what
> colors are most common, medium common, and least common?
>
> Thanks

--------------------------------------------
AE:
Interesting question.  I know that it's known for certain classes of
images.  For example the gamut of skin tones is known to be a
boomerang/banana shape (makes perfect sense once you think about it).
That's just the skin tones though, not the hair or other background.
And I believe that the mean of all colors printed on regular paper is
18% gray (or 12% gray  - http://www.bythom.com/graycards.htm) but I
don't know the probability distribution function of the gamut.  I
think that's the reason that the 18% gray concept was invented.

It may depend on the scale.  If you look at the whole planet, which
theoretically contains every possible image taken on earth, then the
gamut is mostly blue (ocean) and white (clouds) as we've seen from
numerous photos of the earth from spacecrafts.  If you zoom in, then
you may see a lot of green and brown as you start to pick up land and
vegetation on the continents.  If you zoom in even more then the gamut
becomes richer as you start to include colorful man-made objects
(cars, buildings, signs, etc.).

You might try asking in sci.engr.color and rec.photo.digital.

I saw a siggraph paper in 2007 where someone downloaded thousands of
photos somehow from flickr so maybe you can do that and calculate your
own.  They were essentially looking for similar gamuts for database
retrieval by image content, like "show me all photos of a bay" or
"show me all photos of an Italian street scene."

Two interesting Flickr color picker web sites that you'll want to
visit are:
http://color.slightlyblue.com/   Nice and fast response!!!
http://labs.ideeinc.com/multicolr/    Authors say :  "We extracted the
colours from 10 million of the most =93interesting=94 Creative Commons
images on Flickr. Using our visual similarity technology you can
navigate the collection by colour."

I think that the above two web sites should answer your questions
somewhat well (i.e. yes, someone is looking into it).

Regards,
ImageAnalyst

On Fri, 5 Dec 2008 02:57:08 -0800 (PST), AE lover
 wrote:

>Hi guys,
>
>Do you know if there is any paper discussing statistics of colors in
>natural images?
>
>For example, if there is anyone trying to analyze a database of
>natural images and do some statistics to get information about what
>colors are most common, medium common, and least common?
>
>Thanks


Kodak did studies like that many years ago, when they were a dominant
company.

Charles wrote:
> On Fri, 5 Dec 2008 02:57:08 -0800 (PST), AE lover
>  wrote:
> 
>> Hi guys,
>>
>> Do you know if there is any paper discussing statistics of colors in
>> natural images?
>>
>> For example, if there is anyone trying to analyze a database of
>> natural images and do some statistics to get information about what
>> colors are most common, medium common, and least common?
>>
>> Thanks
> 
> 
> Kodak did studies like that many years ago, when they were a dominant
> company.

I wonder if Google has any of that type of data from their image search 
repository.  More importantly if they would release it without an 
exorbitant fee.

Not exactly, but you might find some references here:


Foster, David H, and Kinjiro Amano. 2006. Frequency of metamerism in
natural scenes. Jounal of Optical Society of America A 23, no.
10:2359-2372.
Webster, M A, and J D Mollon. 1997. Adaptation and the color
statistics of natural images. Vision research 37, no. 23:3283-98.

On Sun, 7 Dec 2008 00:38:07 -0800 (PST), oicherman@gmail.com wrote:

> Foster, David H, and Kinjiro Amano. 2006. Frequency of metamerism in
> natural scenes. Jounal of Optical Society of America A 23, no.
> 10:2359-2372.

Thanks.  Interesting - topic.  Found access to a free copy here:
http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=2040061
-- 
Mike Russell - http://www.curvemeister.com