Via #!/bin/ubergoonz -x
I saw this patent filing referenced by the above blog, of an automatic way to detect pornographic images..
While technology comes with its own problems, it also tries to solve its own problems =) A few short comments on the techniques described in the patent claim:
1. I could think of many other images that qualify as pornographic that will not satisfy the criteria listed in the claim (eg: 'skin' areas detected, body shape, possible genitalia shape match, or even *shudder* 'erotic positions'!)
2. Human ingenuinity may test the bound of this flimsy heuristic. One of my computer science professors remarked once that computers/AI of today cannot even recognize whether the object in an image is wet! There is so much capability of our God-given 'wetware brain' that we take for granted, or thought trivial to reproduce/replace by a computer.
3. Most university lab-produced AI system I know so far have only been tested against a small (but academically respectable) number of subjects. Against millions of porn images out there, one may spend ages wading through false positives and false negatives...
This kind of initiative, I suspect, will be a handy first-level 'defence' against pornographic entries that XShare might receive, yet I doubt this algorithm would give satisfactory enough level of accuracy.