Will we ever be able to search the web from images?

[Kermit power]

Look at some of the threads on here. Just now on the front page of the P&P we have a bug and a snake which people have been seeking to identify.

In both those cases, human intervention has quickly provided an answer, but do we think it will ever be possible for a computer to do this? Obviously we already have automated text recognition, and have done for years, but that's within fairly close parameters. Will it ever be possible to upload a picture of something and have a computer identify what it is, based on similarity to other images within a database or something?

[anonymous-user]

Facial recognition technology is advancing every day, I don't see why this couldn't be expanded to other objects one day (if it hasn't already)

[mrmr96]

Yup. They are working on it now. But it's hard. There's a program going around at the moment concerned with differntiating dogs from cats.

[Shabs]

A friend is currently coding real image identification for mobile phones that are internet connected. Simply take a photo of anything and it identifies it - like Shazam for your eyes.

His company have been doing R&D on it for years, not sure when it will be released.

Shabs

[DrTre]

If not by directly analysing an image then with the concept of the "semantic web" and onward, yes certainly, without a doubt.

[SeeFive]

The potential is there to do it - as mentioned face analysis is there today. Simple demonstrations have been written - I used to have one which was a flower database, differentiating flowers by colour, petal shape and overall shape. You could just submit a picture, and it would find a similar flower. For those that know me, I submitted a pictire of me, and it brought back a cactus, which is pretty accurate apart fro the colour.

Also did some stuff recently with aircraft / ship recognition, which is somewhat more difficult as the objects look different depending on which angle you view them from.

The main issue at the moment is the processing power required to analyse all pictorial content on the web in all the various categories of articles (cars, boats, people etc). One of my current customers has a massive picture library online, but only searches them by categories / keyword tags. So, if you have a picture of a "green bug with long legs and whiskers", typing that into a search box would typically return good results. Also, as this type of stuff we sell does some high quality clustering, just typing in "green bug" will return clusters of similarly tagged bugs (e.g., long legs, short legs etc) in their own little category groups. You could of course submit the image, and the parser would strip any tags, and use them to search automatically.

Funnily enough, I spent some time today with a guy who is developing 3D XRay analysis from CTScan machines, automatically identifying organs and issues, allowing drill-down through the body image, into organs, into veins/valves etc. Quite cool stuff, but very specialist, and quite big files to manage across the web, but it's getting there.

So, it is possible by keyword / tags, the image tech is getting better but it would require a lit of tin to do it on the whole internet, and frankly, there is little perceived business need. Semantic web may help.... but don't hold your breath on that one.

[evenflow]

Google are starting down that road.
Go to google.co.uk and select image search.
Search for a person (eg Cindy Crawford)
Click on "show options"
You can then choose to show photos with faces in only

[DaveL485]

There's a phone function....play a clip of music down the phone and it texts you back the song details. Kind of what you are on about but with audio.

So yes, it'll be possible.....and soon, i reckon.

[SeeFive]

evenflow said:

Google are starting down that road.
Go to google.co.uk and select image search.
Search for a person (eg Cindy Crawford)
Click on "show options"
You can then choose to show photos with faces in only

Same thing on http://www.exalead.com/search/image/results/?q=mad... plus a few other metadata oriented options.

[EINSIGN]

Yes:

http://www.youtube.com/watch?gl=GB&hl=en-GB&am...

[mrmr96]

DaveL485 said:

There's a phone function....play a clip of music down the phone and it texts you back the song details. Kind of what you are on about but with audio.

So yes, it'll be possible.....and soon, i reckon.

That's called Shazzam and has been referred to in an earlier post. The chap pointed out that there is something similar being developed for photos. I.e. submit a pic and the service tells you what it is. I don't think this will ever work to such a detailed degree to be useful though.

[TooLateForAName]

There is a service that will find copies of images - developed for finding copyright infringements - that apparently detects even crops and mild edits.

iron something?

Edit - wrong metal, tineye : http://tineye.com/

Edited by TooLateForAName on Friday 25th September 09:03

[The Black Flash]

Google are working on things like that:
http://similar-images.googlelabs.com/

On a slightly separate vein, I'm very impressed with MS OneNote, which has a search that includes text in images. Clever stuff.

[illmonkey]

The Black Flash said:

Google are working on things like that:
http://similar-images.googlelabs.com/

On a slightly separate vein, I'm very impressed with MS OneNote, which has a search that includes text in images. Clever stuff.

That
Is
Awesome

The google bit

[collateral]

evenflow said:

Google are starting down that road.
Go to google.co.uk and select image search.
Search for a person (eg Cindy Crawford)
Click on "show options"
You can then choose to show photos with faces in only

Wonder when it will work with other body parts too

Reminds me a bit of that show James May did with the robot which could identify things. Doubt they will be able to make it very specific because there are just too many variables - telling a car from a table would be easier than telling a '64 Lincoln from a '65

[sonic_2k_uk]

For people in the know, how do you actually go about identifying objects e.g. facial recognition?

I gather the same principles apply to a video or live stream as they do to a picture, and that its a case of progamatically dividing the image up into different sections and analysing the results?

Im a developer, and image recognition software is something that has always interested me despite having absolutely no idea where i'd start....

[Kermit power]

illmonkey said:

The Black Flash said:

Google are working on things like that:
http://similar-images.googlelabs.com/

On a slightly separate vein, I'm very impressed with MS OneNote, which has a search that includes text in images. Clever stuff.

That
Is
Awesome

The google bit

It would be if it let you upload an image of your own to then find similar stuff. If it could do that, it would be exactly what I had in mind. As far as I can see, though, it doesn't let you do that?

[Kermit power]

TooLateForAName said:

There is a service that will find copies of images - developed for finding copyright infringements - that apparently detects even crops and mild edits.

iron something?

Edit - wrong metal, tineye : http://tineye.com/

I tried that. It didn't seem to work.

I downloaded an image from my Flickr page, then uploaded it to there, then clicked search, and it found no results.

[voyds9]

I seem to remember some large group that had software that built peoples photos in to a large montage of say a building so you could walk round it. iirc that was done with software rather than manually, to it was not only able to identify an object but even parts of the object.

[evenflow]

Yep, it's called Microsoft Photosynth, and it's blimmin' clever.

http://photosynth.net/Default.aspx