SDD: A skin detection dataset for training and assessment of human skin classifiers

Recently, skin segmentation has been utilized in wide variety of biometric applications including face detection, recognition, tracking, image filtering, archival and retrieval, etc. Along with its applications, different methods have been designed in order to segment pixels of an arbitrary image into skin and non-skin classes. However, there is no reliable, accurate, appropriate and applicatory dataset either to train or evaluate these algorithms. To this end, a comprehensive dataset, SDD, is introduced in this paper which addresses the limitations of former image libraries. SDD contains more than 20,000 color images accompanying with their manually annotated ground truth. It is suitable for assessment of skin classifiers since it is a very extensive database in which images are divided distinctively (very important from evaluation and training point of view) and it covers multifarious photos captured in all around the world in different conditions. In addition, unlike many other datasets with semi-automatic ground truth labeling, GTs in SDD are very precise thanks to the use of a professional graphical tool and more importantly, the idea of ternary division. The proposed database has been compared to SFA through which both qualitatively and quantitatively, the appealing features of the SDD are confirmed.