I just checked the two examples Rene posted here.
PhotoBucket has marked the image as violating their terms and blocked
it. The other site, Vietnamese, is impossible for me to navigate but I
noticed that the image Tineye posted as being identical was cropped and
is only 4.3kb. I know nothing about the technology. Is there optical
recognition software for images? I would not think it possible to do
that with so many images so quickly. BTW, tineye says they searched
over
1.2769
billion images. No small number but probably just a drop in
the cyber bucket.
Don
On 2/12/10 11:30 AM, PhotoRoy6@xxxxxxx wrote:
I think they are making direct matches. When they say content I
think they mean if you load a picture of a handbag they don't look for
handbags but only for your specific image of a handbag. So they are
searching for a pattern that looks exactly like your pattern (your
image). I assume they are looking at 0 and 1 but there may be other
ways to do this.
Roy
As the number of images drastically increase on the web each
and every year, I doubt there really is any practical way to catch up.
I also wonder one thing. Just how are they making a match? If they
can't see content, what are they using? Surely its more than file name
in that could be changed easily. If they are checking the ones and
zeros, then they probably are looking at content though there may not
have someone looking at content. They may not know its a face, (unless
someone later decided to open the file) but it I guess would be looking
at content.
Now searching for invisible watermarks should be much more
effective, but digimark is still fairly expensive. I hope someone
comes up with a similar service at a much cheaper price.
|
