Boobies and Beaded Headbands
First, the image, via Kinky Delight:
With this image I have my usual problem. After too damn many years of looking at pretty dirty pictures out of context, I find myself wanting to know more about them.
I can’t tell. Are these supposed to be American Indian girls, in some Russ-Meyers-esqe fake-historical western titsploitation drama? Nudists, at a themed naturist camp with a cutesy name like “Lake Hiawatha”? Hippy girls? What?
I have this software fantasy. It involves a distributed app for image searching and collaborative crowdsourced image metadata creation. Sort of a community tagging app for images that exist on the web, with smart checksum-and-fingerprint routines for loosely associating similar images wherever they might be found, and mapping quality metadata from one instance of an image to another. Throw in a cryptographically-secured unique-pseudonyms reputation system and a feedback/voting system to make it easy to identify and filter out vandals and anybody else who pollutes the system with consistently bad metadata. Thumbnails and checksums and metadata are tiny, so making the whole thing person-to-person (and thus, lawyer-resistant) is possible, at some cost in speed and efficiency. Some days I wish I had programming chops.
Shorter URL for sharing: http://www.erosblog.com/?p=5009
I too have fantasized about an App for my ‘dirty pics’ but it’s a bit different in intent. I want the app to first recognize and tag the objects & activities in the pictures (photographic or drawn). Second, I want the app to identify objects with a unique identifier (facial recognition, landmark identification, etc.) And finally third, a link from the database of tags to a web crawler that can continuously search the web for matching objects, automatically downloading said objects.
These are similar goals, but do not involve the voting and other social networking tools.
These would be simple tools. I know there are facial recognition programs, but I have not researched the level of sophistication of the science.
Since I don’t use automatically downloading tools, my personal ‘stash o’ porn’ is only 194 GB, about 114,500 files (95% downloaded from free sites), but the organization of the files into a searchable database is always falling behind.
B
I wish I had the programming chops as well.
For this specific image, one plausible (but not certain) candidate would be a 1969 sexploitation film The Ramrodder.
Odd, aint it, that at some point the faces become more interesting than the bodies. Not that beautiful nude women aren’t interesting, but better a happy smiling amateur than a jaded pouty professional.
Yeah, Earl. I’ve been at that point for quite awhile, too.
There are whole vast swathes of porn that might as well be stock quotes printed on newsprint for me, because it’s unimaginative and even the models look bored with it.
This is fairly interesting, it seems like it breaks down into this (numbered for reference, nothing else):
1) Tag the image (human)
2) Identify duplicates (machine & human)
3) Reputation system (machine & human)
4) Legal issues
What do you mean by “smart checksum-and-fingerprint routines for loosely associating similar images wherever they might be found”? Is this more than a dup checker that handles the same image at different scales?
I’d suggest we build this as a set of loosely coupled tools that cooperate.
Mike, it’s a bit more than a dup checker.
Files that are bytewise identical can be identified as “the same” by a very simple hash or checksum function. Scaling or cropping the image destroys that, but what image search software like TinEye does is create an image “fingerprint” that incorporates unique features of an image and can mechanically spot them at different resolutions and crops. Of course this is not perfect, and it can sometimes identify two different photos of the same scene as “similar” or the same. That’s where the human eye still plays a role.
My thinking is that tagging a single image is a waste of effort; what we want to do is be able to aggregate the tags that different people have applied to different instances of the same (or similar) images. If I’ve got a photo, I want to be able to ask “what tags have people put on other versions of this photo, or on other photos of the same scene, or on other photos that are similar enough to this one?
I don’t really see a way to separate the 1,2,3, functions usefully. Tagging images in isolation on my local client is trivially easy, but the scale of the task makes it pointless. The dupe identification (really, association of similar images together) is an essential part of any community tagging effort, so that a tag I create is not wasted, but is available to people looking at other versions of the image wherever they may be found.
And community tagging will *fail* if there isn’t some system for voting on the tags. There’s got to be a way for people to indicate “that tag is rubbish” and to use those votes to filter the display of tag info. This could be done numerically without a reputation system, but a rep system would allow you to weight inputs — denigrating data by people who rack up a lot of “rubbish tag” votes and enhancing the weights of votes by people whose “rubbish tag” votes are frequently and widely confirmed.
Note that everywhere I say “tagging” I’m actually talking about a few more functions than that. Tagging in its commonest form involves short words or keyword phrases. Although that’s hella useful, I’m interested in a system that would allow more detailed metadata to be entered when known — everything from photographer to publisher to original published source to source URLs.
A color histogram or something along those lines can detect the *same* image at different scales. It could also detect *similar* images from the same series with a bit of fuzzyness thrown in. More than that I’d be dubious about, it sounds like it wanders into image recognition.
I don’t see anything in what you’re writing that precludes a loosely coupled approach, we’d just need to agree on the underlaying syntax & semantics. I’d also think it would afford some amount of legal insulation (I’m not a lawyer nor have I ever slept at a Holidy Inn Express :-)
It *does* wander into image recognition but it can be done to an extent. The boys at TinEye have wandered quite a distance in fact.
Just off the top of my head, I’d say this was taken at a nudist camp in Germany. The Germans have a whole sub-culture about pretending to be Indians.
I believe that pic is from the White House pool during the JFK era.
I tried a reverse image search for it using TinEye.
It came up on FFFFound. I do not like FFFound because there is NEVER any credit to images posted. And the one there, appears to be from here.
Yeah, these days I use TineEye quite a lot. If I express interest in an image, you can pretty much bet I’ve already used TinEye to see what it can tell me. But a source like ffffffffffoundandrippedoffffff doesn’t tell us much. :(
When I read your description of the tool you wanted, it stirred a memory of Tin-Eye, which I see others already discussed. (That saved me the research of having to track it down again. I couldn’t remember the name!)
But even more-so, it reminded me of STIC (Some Tools for Image Collectors), which is an open-source tool collection for Linux that I experimented with a few years ago.
You can find it, and some info, at http://stic.sou....net/ and I think it may have some of what you’re looking for, but in a single-computer app, not a distributed/crowd-source kind of way.
I’ll say up front that STIC was one of those tools that I found which fit into my “hypercomplex” category. (Blender3D is another one, or any really intense software interface, where you have a lot to learn, and you have to KEEP USING IT or risk forgetting how to do all but the simplest of things.)
STIC is somewhat arcane, obscure, I spent a lot of time reading & re-reading the docs, and experimenting with the non-intuitive process. And with a build-up like that, you may wonder why I’m even remembering it some 5 years later? :-)
Because it had some AWESOME and powerful functions. And once you got the hang of it, it actually did an amazing job.
One of the most impressive was recursing through nested image directories with thousands (or tens of thousands) of images, and having it find “similar” things, like duplicates of a different size, different shade/tint/light-level, or those images with a little logo slapped in one corner saying “Copyright of whatever website you got it from.”
Those were the best. STIC would come up with 3-5 images and say “I think these are all the same”, and they are, except that there are 5 different sites claiming copyright! You figure at least 4 of them are lying, and maybe all 5… :-)
I think the thing that impressed me the very most about STIC though was the time it found an image that had been photoshopped in a directory of images from a completely different website that what it was looking at. (It builds its own local database of everything it analyzes.) It took me a minute to figure out what it meant when it said “I think these are similar”, since in one, the woman was blond, and in the other, had brown hair; the faces were different, etc. The thing is, the ‘shopping was done well-enough that you might not notice it was altered, until it came up against what was obviously the original, and then you could see the small details, in among the big changes. (A bit like Tin-Eye does, showing you all the variants.)
I don’t quite know HOW it processed those images to come up with the match, but it did, and earned itself a place in my “suitably impressive” list, which allows me to pass on the recommendation today. :-)
And hey, it’s open-source! So you could probably find some people to help hook it into distributed social networks & crowd sourcing stuff that didn’t exist 5 years ago.
Hope this helps. It’s an interesting toolset, even if it is a steep learning curve… And by building on the ImageMagick tools for analysis, it has a lot of the hard stuff figured out already.
Pat
If you’re still looking for the origin of this photo, you got it right the first time: “Russ-Meyers-esqe fake-historical western titsploitation drama” .. it may even be Russ Meyer, but I’m fairly sure this still is taken from a series run in Playboy in the late sixties called “The history of sex in cinema”, subtitle, “nudy cuties” or some such. I will be able to say precisely which edition of playboy and other data within a few weeks if you’re still interested and hopefully send a few scans.
Ya absolutely, if you find some more scans, I’ll do a post. Thanks!