Pornhub analyzes videos with image recognition to improve tags

Porn site Pornhub has recently been testing software that can analyze videos on the site frame by frame. This makes it possible to automatically add names of actors and put the correct tags with the videos. For example, videos that match the interests of visitors should be easier to find.

The software uses image recognition to recognize characteristics of the video, writes The Next Web. For example, it can recognize faces of actors and actresses, as well as the positions in which they appear in the videos. In addition, it can recognize and group elements such as hair color and decor. In order to recognize the people, Pornhub has pre-loaded official photos of many of the protagonists of the videos on its site into the database.

With the feature, Pornhub wants to be able to automatically add tags that posters of videos have forgotten or misplaced so that users on the site get better search results. The system also checks with users whether the tags and descriptions used are correct. Users can indicate that by giving a thumbs up or down.

In the first test, the system scanned 50,000 videos. Over the next few months, Pornhub wants the software to index all videos on the site frame by frame. That’s about five million. The software should be ready early next year.