Nude.js | Nudity detection with JavaScript and HTMLCanvas

http://www.patrick-wied.at/static/nudejs/

by pa7 • 16 years ago

144 51 16 years ago

snowmaker• 16 years ago

Slightly off topic, but does anyone know of a server-side implementation of the same concept? I've been looking for a good one.

dangrossman• 16 years ago

I wonder how much work it would take to run this server side on node.js. It has a canvas implementation.

pa73• 16 years ago

if anyone goes for a node.js implementation I'd offer my support on it

sandaru1• 16 years ago

There are some commercial products such as http://pikture.logikit.net/Content

and there are lot of research papers if you are interested in implementing them : http://portal.acm.org/citation.cfm?id=1021261 http://www.springerlink.com/content/e774824834226tv2

nailer4• 16 years ago

Yep, I maintain one. https://github.com/mikemaccana/pyfilter

It's a Python server side client for the cheap-but-paid PIFilter web service.

donpark• 16 years ago

combine nude.js with node-canvas at https://github.com/learnboost/node-canvas

Mpdreamz• 16 years ago

Wonder what came first, the clever wordplay or the actual feature. Really cool stuff though.

iwr• 16 years ago

If you opt out of the nude.js, the pat-down bot gets to fondle your dangling pointers.

burgerbrain10• 16 years ago

Now if only they would make one that detected violent and hateful content.

You know, things people should actually be concerned about children seeing.

pa73• 16 years ago

you're right but how would you define violent and hateful content? it's not easy. ok, somebody could build a neural network and train it with typical violent content, but violence has too many types

netcan• 16 years ago

Do you think there is any other type of potentially offensive content that could be filtered in this way?

nailer4• 16 years ago

> you're right but how would you define violent and hateful content?

For one subset of violence, gore pictures, you could identify body parts (already solved) then look for what appears to be wounds using color changes and edges.

That's of course one small part of the concept of violence - you're totally correct in stating this is difficult.

Also hateful is more objective/political so I don't think we have much chance there.

burgerbrain10• 16 years ago

How do you define pornographic content? It's not easy, and this thing doesn't even do it well or cleverly. It just looks for large amounts of flesh. We've had systems that do this since the 90s. This doesn't seem to solve any of the flaws that were present in those older systems, it's only news because somebody coded it up in a popular language.

I would wager it would be rather easy to make a system that detects bloody gore at a higher accuracy rate than this crap. Of course I have little interest in aiding any sort of censorship, so you won't see me attempting it.

sliverstorm6• 16 years ago

It's just a sample reason, nobody says it's the only use case. What if you run library computers or an internet cafe, and don't want customers browsing porn on your machines? (ethics be damned, they are your machines, you make the rules!)

burgerbrain10• 16 years ago

>What if you run library computers or an internet cafe, and don't want customers browsing kittens on your machines? (ethics be damned, they are your machines, you make the rules!)

I'm not seeing either statement as more legitimate than the other. We all know porn is being singled about because of an absurd prudish attitude the west has towards nudity.

confuzatron• 16 years ago

I'm hoping someone will implement tiresomelyRightOn.js.

ben0x539• 16 years ago

I do not know, I see a more profitable or popular use case for a porn detector than for a violence filter.

seanlinmt2• 16 years ago

Interesting concept.Scanning through the code, reading just the comments, it seems to base the nude-not-nude decision by the amount of skin shown.

Hmm.. I guess it depends on what your definition of nudity is.

nailer4• 16 years ago

Indeed, the sample images don't have any edge cases - eg, a small bathing costume, pasties, etc.

Also what about people with unmaintained pubic hair?

Other tools Im familiarmwith do body part detection, which words pretty well, and generates few false positives (that said, when they do occur they tend to be amusing - a pastrami sandwich , for example).

buro9• 16 years ago

Or, a black person.

If the test is based on skin colours, what happens with different tones?

cookiecaper• 16 years ago

Yeah, the code looks like it determines based on what percentage of the image is "skin". The demo page gives a rather close-up shot of the upper-left of a woman's shoulders/chest. What if someone zooms out a little bit? In my tests it is not doing very well if the subject is not taking up the whole image.

Perhaps a better approach would be attempting to characterize certain features, like nipples/areola or buttocks. That definitely sounds like it would require more intensive processing, though, and be highly dependent on the angle of the image.

I don't really know much about image processing and computational analysis, I just like to pretend I know things.

seanlinmt2• 16 years ago

Also, you'll need to figure out if the person in the photo is a girl or guy. :)

seejay• 16 years ago

I couldn't help but click on the message :D

pretty useful concept, i must say...

nailer4• 16 years ago

That makes sense: the girl isn't nude. She has a black area over her nipple. If that's detected as nude, a small bathing costume would also be 'nude'.

matthijs• 16 years ago

Even a really close up picture of a face will be seen as nudity. Trying to detect a body (or parts of a body) and determine how much skin is visible would probably be a better approach.

Also quite relevant is this site: http://www.yangsky.com/researches/physicallinguistics/PLUnde... Besides a "breast detector/nipple detector" he also created a couple of weird detectors like a "cowgirl sex position detector" among others. (The link is sfw, the individual detectors are not)

pa73• 16 years ago

The algorithm is mostly based on this paper: http://www.math.admu.edu.ph/~raf/pcsc05/proceedings/AI4.pdf but there are some steps open. I've implemented this algorithm because it's not as hardware intensive as the usual nude detection algorithms (such as searching for specific body parts)

cabalamat2• 16 years ago

> The detection algorithm runs at the client

Thank you. I really want Firefox to be even slower running unnecessary JavaScript.

burgerbrain10• 16 years ago

I'm not sure why this is downvoted. It runs really slow on firefox...

cabalamat2• 16 years ago

Sadly, everything runs real slow on Firefox. I'd be using Chrome if it had Ad Block Plus.

Nude.js | Nudity detection with JavaScript and HTMLCanvas

Related Stories