Negative content filtering based on skin texture, homomorphic filter and localizations

There are many challenges in the development of negative content filtering system. One of the techniques used in negative content filtering systems is by using skin segmentation. There are two challenges in skin segmentation. The first is skin-color similarity problems such as sands, woods, and animals which have similar color with skin and the second is low light intensity problems. In this research, the data for training and testing are divided into two. First data contain porn category, skin category, and non skin category. These data are trained and tested using GLCM textures to classify skin. The data of skin area is then compared to localization area and its whole image to classify porn category. Second data contain porn image in low light intensity. These data are trained and tested with homomorphic filtering. Testing on the first data shows that the accuracy of porn detection in porn, skin, and non skin categories are 95.5%, 93%, and 98.7% respectively. Whilst testing on the second data shows that the accuracy of porn detection is increased by factor of 3.38 when homomorphic filtering is applied.