Pornography Detection Using Support Vector Machine
Yu‐Chun Lin, Hung‐Wei Tseng, Chiou‐Shann Fuh · 2003
As Internet grows quickly, pornography, which is often printed into a small quantity of publication in the past, becomes one of the highly distributed information over Internet. However, pornography may be harmful to children, and may affect the efficiency of workers. In this paper, we design an easy scheme for detecting pornography. We exploit primitive information from pornography and use this knowledge for determining whether a given photo belongs to pornography or not. In the beginning, we extract skin region from photos, and find out the correlation in skin region and non-skin region. Then, we use these correlations as the input of support vector machine (SVM), an excellent tool for classification with learning abilities. After a period of training SVM model, we achieved about 75% of accuracy, 35% of false alarm rate, and only 14% of mis-detection rate. Moreover, we also provide a simple tool based on our scheme.