How to programmatically identify sexually explicit images? [closed]

Question

Closed. This question needs to be more focused. It is not currently accepting answers.

---

Want to improve this question? Update the question so it focuses on one problem only by editing this post.

Closed 8 years ago.

Improve this question

Google does this successfully. Do they use an algorithm or are the images marked manually by people once and the results are used by the exact same images thus significantly reducing the number of images to go through?

Because I am not sure how else you can identify these pictures. You can use skin color but is it 100% accurate? What if the person just wore a skin colored outfit? Or are these eliminated too?

Or what if the person is fully clothed but performing sexual acts?

Answer 1

It is just another image classification or object recognition problem. Searching on Scholar.Google for Pornographic image is a great start. Neural networks would be my first attempt (and seems to be a popular choice from the pubs on Scholar).

This paper, Recognition of Pornographic Web Pages by Classifying Texts and Images, has a good related works section that you could investigate. It lists model-based, region-based, and feature-based as the main techniques to analyze images for pornographic content.

Your idea of using the skin color is very naive and wouldn't be very accurate (would be interesting to use as a comparison for a more sophisticated method though!).

If I was to try to tackle the problem, I would start with Viola-Jones objects (based on Haar cascades), which is already implemented for you in OpenCV. Also read up on neural networks. I have used these techniques to successfully find particular muscles in images of people so there is no reason why you can't look for explicit content with them. (Disclaimer: This is not my field of research, only a hobby, so make sure you go into those papers!)