Google and Watermarks

It looks like Google is headed to the darkside.

https://www.gizmodo.com.au/2017/08/g...matically-and-
perfectly-erase-watermarks-on-stock-photos/
http://tinyurl.com/y88mrfvx

--

Regards,
Savageduck

On Aug 19, 2017, Savageduck wrote
(in iganews.com):

It looks like Google is headed to the darkside.

https://www.gizmodo.com.au/2017/08/g...matically-and-
perfectly-erase-watermarks-on-stock-photos/
http://tinyurl.com/y88mrfvx

However, they also supply a way to circumvent their WM removal method.

--

Regards,
Savageduck

In article .com,
Savageduck wrote:

On Aug 19, 2017, Savageduck wrote
(in iganews.com):

It looks like Google is headed to the darkside.

https://www.gizmodo.com.au/2017/08/g...matically-and-
perfectly-erase-watermarks-on-stock-photos/
http://tinyurl.com/y88mrfvx

However, they also supply a way to circumvent their WM removal method.

I wouldn't trust them on that latter claim...
--
teleportation kills

"android" wrote

| However, they also supply a way to circumvent their WM removal method.
|
| I wouldn't trust them on that latter claim...


The man who tells you your lock is broken
is not the thief.

And they're not "claiming" anything. They reported
that they discovered it's possible to identify so-called
digital "watermarks".

(I find watermark to be a misleading term here. A
watermark is a hidden ID. Digital marks are glaring
blemishes that make the image useless. A digital
watermark would be hidden, unique pixels that allow
the image to be ID'd. I think that method has actually
been used with music files.)

Google also noted that if the mark is varied then
the task of finding it becomes much more difficult.

Even if someone goes to all the trouble to remove
the marks, there's no way to recover the missing
pixels, except by something like a "smart fill"
function. That's an awfully lot of work, just to end
up with a mediocre image that can't legally be used
commercially. (Note that Google's sample image is
of a soccer game. Most of the mark is over nearly
solid color blocks: black socks, green grass, etc.)

Someone who wants a stock photo can probably
find a legal image for free that's good enough. If not,
they'll buy it. For the most part it's companies who buy
stock photos, for things like ads or catalogues. Those
companies are not going to take legal risks by using illegal
photos. It's just not worth it. In other words, if you're
currently making money selling stock photos, this news
is not likely to ever be a threat to your income.

So it's more a curiosity than an issue.

If you're still concerned, you can edit your photo
before putting it online, adding dots or lines to the
mark to make it slightly varied.

In the end, you face a much more important problem.
The same dilemma that nearly all artists face: Most art
just isn't worth much in terms of money.

In article , Mayayana
wrote:


Even if someone goes to all the trouble to remove
the marks, there's no way to recover the missing
pixels, except by something like a "smart fill"
function. That's an awfully lot of work, just to end
up with a mediocre image that can't legally be used
commercially. (Note that Google's sample image is
of a soccer game. Most of the mark is over nearly
solid color blocks: black socks, green grass, etc.)

the computer does the work for you, with very good results.

at google i/o a few months ago, google demoed removing a chain link
fence, entirely automatically.

https://www.theverge.com/2017/5/17/1...-object-remova
l-feature-demo

In article ,
"Mayayana" wrote:

Google also noted that if the mark is varied then
the task of finding it becomes much more difficult.

From the linked article:

"The company instead took this research one step further and found that
by slightly warping a standard watermark from image to image, even the
smallest of deformations made it impossible to automatically remove the
watermark without leaving obvious artefacts behind."

The article could be wrong but it's author claims that Google have ditto
that automatic removal of watermarks can be prevented. I just wouldn't
trust them on that...

https://www.gizmodo.com.au/2017/08/g...utomatically-a
nd-perfectly-erase-watermarks-on-stock-photos/
--
teleportation kills

"nospam" wrote
|
| at google i/o a few months ago, google demoed removing a chain link
| fence, entirely automatically.
|
| https://www.theverge.com/2017/5/17/1...-object-remova
| l-feature-demo

I can only see the chain link. They've gone to great lengths
to force the visitor to enable javascript.

In any case, that's another example of the same thing.
The chain link fence consists of thin lines with greenery
in the background. The greenery is essentially just a random
pattern of greens. So all they need to do is fill in the pattern
with similar areas of greens in the same range.

I'd be more impressed if they showed a variety of images.
For instance, maybe something like this...

http://www.istockphoto.com/photo/cof...97744-81416497

.... where the exact color of removed areas will be important
to the image.

Or even this...

http://www.istockphoto.com/photo/del...76971-42523300

.... where gradients, regular patterns, and small areas of
extreme color change are important.

"Mayayana" wrote

Ah. I found the original. Interesting. That
is impressive. Except for a faint, darker area
going diagonally across the boy's shirt, I can't
find any trace of the foreground fence. I guess
maybe ID marks should be made wider, so that
it's harder to fill in the background.

In article , Mayayana
wrote:

|
| at google i/o a few months ago, google demoed removing a chain link
| fence, entirely automatically.
|
| https://www.theverge.com/2017/5/17/1...-object-remova
| l-feature-demo

I can only see the chain link. They've gone to great lengths
to force the visitor to enable javascript.

your loss, and it's just an animated gif anyway. it's completely
harmless.

https://cdn.vox-cdn.com/uploads/chor...59/May_17_2017
_13_15_18.0.gif

In any case, that's another example of the same thing.
The chain link fence consists of thin lines with greenery
in the background. The greenery is essentially just a random
pattern of greens. So all they need to do is fill in the pattern
with similar areas of greens in the same range.

it's a lot more complex than that.

try doing it yourself, manually, *without* content-aware fill (which
you most likely don't have anyway).

just do what you say is needed: fill in the pattern with similar areas
of greens, then post the results.

I'd be more impressed if they showed a variety of images.

you underestimate what machine learning can do.

On 20/08/2017 15:08, Mayayana wrote:
"nospam" wrote
|
| at google i/o a few months ago, google demoed removing a chain link
| fence, entirely automatically.
|
| https://www.theverge.com/2017/5/17/1...-object-remova
| l-feature-demo

I can only see the chain link. They've gone to great lengths
to force the visitor to enable javascript.

In any case, that's another example of the same thing.
The chain link fence consists of thin lines with greenery
in the background. The greenery is essentially just a random
pattern of greens. So all they need to do is fill in the pattern
with similar areas of greens in the same range.

I'd be more impressed if they showed a variety of images.
For instance, maybe something like this...

http://www.istockphoto.com/photo/cof...97744-81416497

... where the exact color of removed areas will be important
to the image.

Or even this...

http://www.istockphoto.com/photo/del...76971-42523300

... where gradients, regular patterns, and small areas of
extreme color change are important.

Regardless, that is OLD news!

https://www.theverge.com/2017/5/17/15654476/google-photos-object-removal-feature-demo


--
“Men occasionally stumble over the truth, but most of them pick
themselves up and hurry off as if nothing ever happened.� (Winston S.
Churchill)