Julian Stöttinger

Skin Colors

Contextual Adaptive Skin Detection

User generated video content has become increasingly popular, with a large number of internet video sharing portals appearing. Many portals wish to rapidly find and remove objectionable material from the uploaded videos. This paper considers the flagging of uploaded videos as potentially objectionable due to sexual content of an adult nature. Such videos are often characterized by the presence of a large amount of skin, although other scenes, such as close-ups of faces, also satisfy this criterion. The main contribution of this paper is to introduce to this task two uses of contextual information in the form of detected faces. The first is to use a combination of different face detectors to ad- just the parameters of the skin detection model. The second is through the summarization of a video in the form of a path in a skin-face plot. This plot allows potentially objectionable segments of videos to be found, while ignoring segments containing close-ups of faces. The proposed approach runs in real-time. Experiments are done on per pixel annotated and challenging online videos from an online service provider to prove our approach. Large scale experiments are carried out on 200 popular public video clips from web platforms. These are chosen from the community (top-rated) and cover a large variety of different skin-colors, illuminations, image quality and difficulty levels. We find a compact and reliable representation for videos to flag suspicious content efficiently.

Skin video database with ground truth

Evaluation of our skin detection approaches is mainly done on our dataset annotated by Christian Liensberger.

Download it here.

more info at

Skin Paths for Contextual Flagging Adult Video Julian Stöttinger, Allan Hanbury, Christian Liensberger, Rehanulla Khan, Proceedings of the 5th International Symposium on Visual Computing (ISVC), Las Vegas, NV, Nov 30 - Dec 2, 2009.

Skin Detection Through Face Detection

This work considers the flagging of user-uploaded videos as potentially objectionable. The main contribution of this work is to introduce two uses of con-textual information in the form of detected faces. The first is to use tracked facesto adjust the parameters of the skin detection model. We develop classification rules based upon a prior face detection using the well known approach from Viola et al. This work builds on our previous work [Khan et al., Liensberger et al.] where we show that more precise adaptive color models outperform more general static models especially for reducing the high number of false positive detections. Humans need contextual information to interpret skin color correctly. We extend the approach by using a combination of face detectors: We combine frontal face detection and profile face detection in a combined tracking approach for more contextual information in the skin color representation. The second use of face information is through the summarization of a video in the form of a path in a skin-face plot. This plot allows potentially objectionable segments of videos to be extracted, while ignoring segments containing close-ups of faces. We show that the properties of the skin paths give a reliable representation of the nature of videos. The proposed approach was kept algorithmically simple, and currently runs at over 30 frames per second. A high level of performance is required in such an application to cope with the large number of uploaded videos.