Our database now has over 140k images and it growing everyday. We are starting to fill out some of the "not-so-popular" keyword areas and grab some speciality images. I am always suprised at some of the great images that come in and are being sold on these affordable agencies.

Image Content Filtering (Nudity, Violence, etc, etc, etc)

I am starting work on the content filtering system. I have decided to seperate the safe version of imageTrail onto a site called www.safepix.net (Family Safe Image Search) I did this so that organizations and families that want to "force" the use of safe searching can check based on the domain name. I will be allowing content filtering here, but inforcing it on the other domain.

The Nude-O-Meter

I am trying to create a methodology to detect nudity with some degree of accuracy with out having to stare at the pictures. I am thinking about implementing a system much like SPAM filters use to check email messages. It would be a scoring system that attaches weights to certain words found in the title or keywords. Enough "nudy" type words and the image gets filtered. I am going to have to run a lot of tests and tune the system quite a bit to minimize false positives, but I think it can work.

I have thought about creating a test set of dat (10,000 images or so) and running the filter on them. I will have hand checked the images and then I can have the program decide what the best weights are for each for set. Hmmm.... Not sure....