Implementation Of Network Pornographic Image Filtering System Based On Ratio Feature

While bringing resource sharing to people, the internet also widely spread harmful information such as violence, reactionary and especially pornography, which seriously affect the physical and mental health of teenagers. According to statistics, new internet images are more than80billion every year and quite a big part of them are pornographic images. Therefore internet content security has become an important part of information security. More an more people are concerned about how to detect and filter naked pornographic images in order to purify the internet environment.In this paper, we design a network harmful images filtering system, which is divided into two modules--webpage processing module and image processing module. The former is used to fetch webpage images and to shield the webpage which contains harmful images according to the latter. The latter is used to analyze images fetched by the former and to judge whether there is harmful images in real time.For the webpage processing module, we make use of the IShell Windows interface of the windows OS shell. Firstly we create a IShellWindowsPtr interface variable, then use the interface variable to traverse and get all the IE instances. After that, we can use related functions of IE smart pointer--SHDocVw::IWebBrowser2Ptr to process the webpage such as to fetch webpage content or to redirect.For the image processing module, comparing to image retrieval task in other fields, It has strict requirements in two aspects. One is high precision, because false negative will seriously reduce the effect of the filtering system and false positive will shield users’normal information and violate users’legitimate rights. The other one is high instantaneity, we should achieve online monitoring and filtering without disturbing the normal operation of the network. Therefore, the technology of harmful images detection keeps making improvement and optimization.Nowadays among most of the network pornographic image filtering methods, exposed skin ratio is still frequently used. However, it has some defects such as the interference of analogous skin color, the interference of non-key-area skin and sometimes the irrelevance of whole exposed skin ratio and exposed extent. This algorithm make use of the consistence of skin color and human body structure, calculate face size ratio, whole exposed skin ratio whose skin color model has been improved by face skin color data and key-area exposed skin ratio. Problems above has been effectively solved by using thens three ratio features. Besides, add a process of turning image to the face detection module, so it can detect severe inclined face effectively. Experimental results show that the recognition rate can achieve91.2%.