h a l f b a k e r y[marked-for-tagline]
add, search, annotate, link, view, overview, recent, by name, random
news, help, about, links, report a problem
browse anonymously,
or get an account
and write.
register,
|
|
|
Now we have mobile phones with cameras and web access, it is time for a database of objects that can be searched by image. There are visual dictionaries available, but I'm thinking more of a continuously updated base of products and parts of products so that they can be identified quickly.
This could
be useful for explaining what the broken bit is to a sniffy car parts guy, or seeing if that thing you've just partially dug up is a 1985 Chevy muffler or a land mine.
I don't doubt that there are SF writers who have thought of this, but I think we must be nearly at a point where it could be practical to do.
As product liability cases get more and more stupid, there could be compulsion to do something like this from Government, after all we are already marking stuff for recycling and safe disposal.
Note: [entremanure] 's dealybobber field guide should be part of this.
One of the obvious problems is scaling, as it is difficult to tell from an image whether it is a large one far off, or a small one near to, but this could be overcome as the base would return a list for you to choose from with approximate sizes.
Its for flowers but the same concept.
http://orchidspecies.com/index.htm flowers flowers flowers? [toomer34, Aug 21 2005]
3D search engine
http://shape.cs.princeton.edu/search.html The first online one, but not the only one any more... [moomintroll, Aug 22 2005]
3D shape from photos
http://www.eng.cam..../digital_pygmalion/ tools like this have existed for a while, but this is the state of the art (I think) [moomintroll, Aug 22 2005]
[link]
|
|
i can see father ted esque no this is small that is far away arguments, |
|
|
There's certainly no shortage of academic and commercial work on image recognition and classification. Image recognition systems can be used for things like recognising faces, detecting blemished fruits, text recognition, and recognition of parts in a factory environment. The MPEG 7 standard is designed in part to allow images to be indexed by a visual description of their contents. |
|
|
The only real problem is the amount of images (and processing time, since that will scale in part by database size): acquiring, processing and storing them. It's primarily an issue of photographing everything in the world, rather than the need for any special new technology. |
|
|
Also, obviously, even trained humans can't recognise just anything from a low-resolution cameraphone image (sometimes it's hard to recognise your friends faces, let alone anything stranger). At the end of the day, it's going to be impossible to identify a lot of things no matter what's doing the recognition. |
|
|
I see, it's a visual database of *everything*. Ambitious to say the least. |
|
|
Is this for 2D images of things or 3D representations of them? In any case, any automated system will need some context, but there's no reason you can't say, "It's a face; tell me who it is", or "It's a bit of my car; what is it?" alternatively, you take several photos and post them; the system reconstitutes this into some sort of 3D shape then looks for that. This is getting very close to being technically feasible, and should be infinitely better, since a database of images is going to get confused by slight changes in camera angle, lighting levels, etc. |
|
|
[Ian Tindale] may be luckily wrong
about one thing: "There's no real way of
saying to the internet, 'What are those
Ray-Bans I'm wearing in this picture?'"... |
|
|
I'l bet that there's a place on the 'net
where people talk about Ray-Bans -- a
Ray-Ban online community where
Ray-Ban geeks would happily argue
incessantly among themselves about
whether those were 1989 Aviator II's or
1988 Surfer Specials (or whatever). |
|
|
There's an online community for pretty
much everything these days, so there
may very well be a place to get those
precious sunglasses identitified. |
|
|
What I had in mind was a more sophisticatred system where you could have uploaded your picture and had the database match the image - but using the old wetware method worked for you, and I'm pleased you made it. Ebay must have the most diverse image collection anywhere by now. |
|
|
Perhaps that would have been a job for the Mechanical Turk or some other kind of outsourcing service, had they been available in 2007. Cheap humans performing better than machines. |
|
|
Seeing as how I'm one of those self-taught mechanics who
walks into NAPA, tosses a greasy chunk of metal on the
counter, and says "I have no idea what this thing is called,
but I need a new one," I would definitely make frequent
use of a VTD. [+] |
|
| |