Forum Moderators: open

Message Too Old, No Replies

theories of whats to come in search

         

ritualcoffee

8:29 pm on Jun 24, 2003 (gmt 0)

10+ Year Member



Imagine if you could search on an image - using an image, or search for voice recordings using a voice file...

[physicspost.com...]

Now, throw out your wild ideas

SlowMove

8:40 pm on Jun 24, 2003 (gmt 0)

WebmasterWorld Senior Member 10+ Year Member



The author seems to be talking about things at a very high level. I couldn't imagine how difficult coding something like that to run fast would be, not to mention the costs of the hardware. I'm sure it's possible, but I doubt if it will happen any time soon.

ritualcoffee

12:25 pm on Jun 25, 2003 (gmt 0)

10+ Year Member



I completely agree...but the idea is wonderful.

Glacai

2:18 pm on Jun 25, 2003 (gmt 0)

10+ Year Member



Excellent idea and definately something we'll see in the future (well we might not) once pattern matching is up to it. I read a tutorial on ocr using neurals nets that would be interesting to try. Then the images could be crawled, analysed and indexed somehow, making only the initial searched image analysis slow, plus the size of graphics would mean a large index! Beyond me but interesting stuff.

msgraph

2:27 pm on Jun 25, 2003 (gmt 0)

WebmasterWorld Senior Member 10+ Year Member



Interesting article.

Some of what he says sounds a little far-fetched yet I believe part of it is reachable sometime in the future.

There is already software out there or being further developed to detect images with large amounts of skin tones to help determine if they are adult in nature.

Who knows what else is currently out there or being developed, especially by government agencies. For example, the National Security Agency had a patent issued this year on a method of extracting text present in a color image [patft.uspto.gov]

I can't see why a search engine should not be able to program ways to detect horizons, buildings, nature, maps, abstract etc. and try to find similar matches based on an image you input into the system. For example, you scan in a colored image of a snow-capped mountain with clear sky all around, then the system outputs a load of similar images. Mix in some user input and AI and after a while patterns match to where text classification can be tagged to the images that have similar color schemes. Of course there will be many flaws that will take time to weed out, but at least something of this sort can narrow down items on a grand scale, yet at the same time not displaying good choices that were mistakenly mis-matched.

Now I think what we'd all like to see is a search system where you can hum or mumble that tune you heard somewhere yet don't know the lyrics, band, or song title. :)

TGecho

7:12 pm on Jun 27, 2003 (gmt 0)

10+ Year Member



"Why do I keep getting penalized in these pine tree searches?"
"Cause you've got a whole mountain covered with em. Google thinks you're trying to cheat."