Augmentation of Skin Segmentation

Skin detection is used in applications ranging from face detection, tracking body parts and hand gesture analysis, to retrieval and blocking objectionable content. We present a systematic approach for skin segmentation with graph cuts by using local skin information, universal skin information and skin augmentation using the off-line learned model. The skin segmentation process starts by exploiting the local skin information of detected faces. The detected faces are used as foreground seeds for calculating the foreground weights of the graph. If local skin information is not available, we opt for a highly adaptive universal seed. To increase the robustness we learn a decision tree based classifier. The learned model is used to augment the universal seed for skin segmentation when no local information is available from the image. Experiments on a database of 8991 images with annotated pixel-level ground truth show that the systematic skin augmentation approach outperforms universal seed only approach, universal seed plus face detection approach and the learned decision tree based classifier (J48), showing its suitability for robust skin segmentation. We also report the behavior of a skin detection system in the presence of skin only images and images which do not contain skin at all.