Skin detection based on local representation of YCbCr color moment

Information distribution that is getting easy and fast causes anxiety among most people in the society, especially about the information containing inappropriate content for particular age groups, for example the pornographic content. The large amount of accessible data requires a system that is able to automatically detect the occurrence of pornography in an image or video. Detecting pornographic image can be done with skin detection process. This research is aimed to detect skin based on local representation from color moment (mean, standard deviation and variant) YCbCr color space. The method used for the classification is Support Vector Machine with Radial Basis Function kernel. The test will be done using 10-fold cross validation toward 400 images consisting of 200 skin images and 200 non-skin images, furthermore comparing the accuracy result of local representation and global representation for the whole image. The result of the test will show that local representation is more accurate compared to global representation. On the other hand, feature extraction that has the most ability to represent skin color is the YCbCr color space with color moment consisting of a group of mean, standard deviation and variant.