A method of erotic images filtering in real internet

For the actual demand of images filtering on the Internet, we present a new method of erotic image filtering in real Internet environment with asymmetric data set. By constructing the mixture skin color model in different color space to detect naked skin color regions, using Adaboost algorithm to detect the human face, analyzing the typical geometrical structure of naked skin color regions, estimating and analyzing the background of inspected images, the method can improve the average recognition rate effectively. In the real Internet environment that the ratio of normal images to erotic images is 40:1, experimental results show that the recognition rate of erotic images can achieve 64.2%, the recognition of normal images can achieve 96.3%, the average recognition rate can achieve 95.5%.