Adult Content Detection and Filtering
The growing number of people and content online makes content filtering, especially for adult content (and specifically pornographic videos), a key issue. Kuznech developed a practical and scalable solution that uses scene classification, logo detection, skin detection, and several other methods to detect pornographic images and videos.
- Scenes & Objects Classification
Kuznech processed hundreds of hours of porn videos, identifying more than 30 of the most typical and frequent explicit scenes and objects. They include images of genitals, their combinations, typical copulation scenes, different sexual positions shooted in a variety of camera angles. With the help of convolutional neural networks Kuznech Content Filtering finds and recognizes these scenes and objects, using those results to mark content as prohibited.
- Logotype Detection and Recognition
Porn studios use a logo or name to mark their videos, giving Kuznech “signals” that those videos may contain sexual (pornographic) scenes.
- Warning Text Recognition
The law requires legal porn studios to place a textual warning screen at the beginning of their videos to inform viewers that they contain sexual scenes. Kuznech compiled keywords (or key phrases) frequently used by pornographers (such as “sexually explicit conduct”), treating them as markers for prohibited content.
- Skin Detection
Most adult films show people undressed to some degree during most of their sequences, and therefore can be differentiated from non-adult content by recognizing the areas showing skin in a representative set of frames. The same can be applied to images.
- Face Search and Recognition
With the help of its advanced Face Detection Technology, Kuznech analyzes content to detect human faces and identifies porn actors/actresses registered in its database. Once a porn actor has been detected, that information is added to the algorithms in use.