Research On Feature Vector-Based Pornographic Image Detection

With the fast development of science and technology, the Internet is getting much closer to us. Surely it benefits us a lot, but at the same time, the pornographic picture spreads wildly in the Internet. The pornographic picture does much bad effects to Internet users, especially to those young people. So it is necessary to develop a eroticism filtering system. By now, there are some Internet- eroticism filter systems, but they don't work effectively. It is not enough to filter eroticisms by blockading Web addresses and filtering erotic texts. It is necessary to use computer vision technique to analyze the images. In this paper, we study the computer detection technique and present a pornographic picture detection technology based on the feature vector of picture.The presented system changes the image size firstly to fast the speed of following algorithms. To alleviating the effects of lighting and some other factors, the images should be lighting-compensated. Skin detection is very important in pornographic image filtering system. Here we detect the skin pixel by the pixels's color in certain color space, then we analyze the texture of the corresponding skin region in skin color masks. By doing this, we can wipe off some noises. We extract features of skin clolor mask images and put these feature vectors to Support Vector Machine(SVM). SVM has two steps. One is learning step, the other is test step. In learning step, we put the feature vectors obtained from the images pre-clsssified manually into SVM to train it. In test step, feature vector of image is put into learned SVM. Whether the image is porngraphic one or not can be classified by the output of SVM.The results show that the Feature Vector-Based Porngraphic Image Dectection Technology works efficiently.