[NBLUG/talk] Searching for near-identical photos?

Aaron G nite at sonic.net
Mon Nov 7 20:11:08 PST 2005


Bill Kendrick wrote:

>On Tue, Nov 08, 2005 at 02:43:36AM -0800, Aaron G wrote:
>  
>
>>That seems like it would be EXTREAMLY hard for a computer to do...
>>The only way I could see it possibly working is that you had some kinda
>>metric for "probility the picture is a match" and you could define your
>>threshold.
>>    
>>
>
>Well, hard or not, KDE gives me an option when I right-click an image:
>"Search for Similar Images..."
>  
>
It means.. search for images of a simliar type or something, I highly
doubt it would mean search for images of a similar object. "find all
pictures of my cousin bob" ... that just won't work.
Think about how an image is stored on the computer.

>It appears to use the image index server (which I know nothing about),
>which is accessible from the KDE Control Center under
>"System Administration" -> "Image Index".
>
>This might be a GUI frontend to some command-line- and server-based tools...
>
>
>  
>
>>This is the same kinda thing as teaching a computer to catch a baseball,
>>or catching it yourself.
>>    
>>
>
>Well, it _is_ the 21st century.  My computer can talk, check my spelling and
>grammar, connect me to thousands of people all over the world, etc. etc. :^)
>
>  
>
Yes.. It is the 21st century, but "computers" have only been around for
about 50 years, plus all of those above things are pretty easy to do.
Computers are bad with recognizing patters, whereas humans are pretty
good, this is part of what allows us to recognise people in dark light,
far away, partial images, that kinda thing... to a computer, this is
VERY challanging. (ie. why facial recongnition software sucks).
The same kinda thing applys to your picture problem.

 -Aaron




More information about the talk mailing list