Google safesearch basically removes NSFW images from an image search. It seems to work extremely well. How do they actually go about doing this?
I assume the best way is to get a large number of people marking images in some way as they come across them, possibly as some sort of browser toolbar. There are so many images out there that it seems cost inefficient for google to have a large staff dedicated to it.
Or is it done heuristically? Does goggle have nipple detector technology? Do they look at the context of the page to determine likely the content of the image? It seems like a heuristic method would have a significant rate of failure, but I can’t recall ever seeing a NSFW pic with safesearch turned on.
It is done heuristically. The primary criteria for excluding images are context and the color content.
Images which appear on pages with lots of text talking about what a filthy cocksucking whore you’re looking at are good candidates. Other good candidates are images with a large percentage of fleshtone colors. Combining those two into some sort of scoring system with a tweakable threshold does a pretty good job eliminating most NSFW images.
Of course, Google needs to take context into account to do image searching in the first place. If an image would be returned in searches for sex-related terms, then it should probably be excluded from a Safe Search.
Do you know for sure that it is looking at the images? I had assumed it was implemented based primarily on URL and secondarily on keywords. It seems redundant to process the images after that.
“One experiment is worth a thousand expert opinions.” (Thomas Edison) Someone should take a scathingly pornographic photo, tint it blue in Photoshop, and post it on a page with lots of text about angels and bunnies and cupcakes. Then see if Google can cough it up!..TRM
Which is pretty disappointing. I mean, that would be a totally sweet job:
“Safe…Safe…Ehhh–Safe-ish. Not really showing anything…Safe…Oh, yeah! Mmmm-mmm–Totally not safe. Oh yeahhh–OK. OK. Back to work…Safe…Safe…Safe…Oh my God–look at the SIZE of those things! Not even REMOTELY safe!”
Although I suppose you’d have to screen some pretty :eek:-inducing stuff.