Hey .. I had an issue last weekend where it seemed the thumbnails were going screwy. It would deliver porn thumbnails instead of the real ones.
http://www.yald.com/technology/0504yahoopornthumbnail.html
|
Such an issue is (theoritically) fairly simple to fix. All a binary process needs to do is study the amount of pixels which are similar to each other in colour within a pre-defined limit. For example, a computer program should look at a picture and see that there are a group of 60x40 pixels which are all between dark pink and light pink and are close together. This would tell you that you're either looking at skin, or somebody's pink shirt – the difficult part is working out what is pr0n and what isn't. Google seem to have it nailed though, unless they're paying people to sift through millions of pictures each month. |
I agree it's not simple. If I enter "Bill Clinton", I do want portraits and such which include a lot of skin tones. I do not want porn, which includes a lot of skin tones. Just see this safe-search for Bill Clinton on Yahoo (incidentally not safe for work): http://images.search.yahoo.com/search/images?p=bill+clinton&sm=Yahoo%21+Search&fr=FP-tab-img-t&toggle=1 |
...unless theyre paying people to sift through millions of pictures each month...
How would one apply for that job?
|
I figured that Google filtered out adult images by analysing the content around the image (and / or the pages / domains linking to the images) for "adult" terms. I know it wouldn't be 100% successful, but surely that's the easiest way to do it.
Having said that, it wouldn't surprise me if Google did pay people to sift through millions of pictures each month. Maybe that's why their image search takes so long to update...
Does anyone actually know how Google filter out adult content from their web or image searches? |
Tony, I think the same, at least this ought to be one of the approaches in the mix. It's not Google's style to hand-pick results, but they might make an exception in web and image search when it comes to very popular 1-term "test" searches. For example, when you enter a certain animal name which is also the slang name of a sexual reproductive organ, Yahoo completely blocks results (asking you to turn off safesearch), while Google shows only the animals in safesearch (and it shows the organ only when safe-search is off). Yahoo certainly seems to hand-pick the blacklist of terms... but I wonder if Google handpicks the images for such search terms. |
Don't the big 3 using some sort of vecter/raster graphic analyis ? Super imposed on graphic content is the psuedo meta data of skin ton, posture balance etc. Linked in with known ip /userprofiles, it then churns out exception that then is tagged for "adult" content safe. Ansd then it goes thru the human filter ?? |