The Research And Implementation Of Network Erotic Image Detection System

With the rapid development of the internet, all kinds of sensitive information including pornographic content wantonly spread. It has seriously interfered with people’s normal online life, especially for young people’s mental health. Therefore, providing neat and clean network environment has become an important responsibility of the network authorities. The existing monitoring filtration systems’work style is to monitor the erotic image information in the terminal which leads to many problems such software installation and operation. So developing a software which can monitor the data source and prevent the erotic image from spreading on the internet is the most suitable way.In this thesis, an IDC-backbone-based erotic image defection system is designed and implemented, which can head off the erotic image in the data source. The main work of the thesis includes:1. The slot-data-package-grouping based image recovery technology is researched and implemented. We allocate the file space in the memory in advance and then put the TCP package to the right position according to the serial number of TCP package to finish the TCP session restructuring and group the image quickly.2. The HASH-based URL duplicate-removing operation and the MD5-based image duplicate-removing operation are researched and implemented,After getting the URL, we do the duplicate-remove operation in the memory with the HASH method, and after getting the image, we do the duplicate-removing operation in the memory with the MD5method. The duplicate-removing operation reduces the image number of downloading and detecting obviously.3. A support vector machine based erotic image detecting algorithm is designed, the algorithm combines the skin region features and face features detecting. We extract the color, texture and geometry features from the skin region and extract the number and the area feature from the face detection result to combine the final feature vector. At last a trained support vector machine will give the judge result. 4. The erotic image detection model for specific categories images is designed and implemented. A fast detection algorithm for some special erotic images such naked-talk or prostitution images is designed. In this algorithm, we combine the local text detection with special position skin detection and detect the naked-talk and prostitution images effectively. In the text detection procedure, we use the integrated image to fasten the process speed greatly.The IDC backbone network erotic image detection system is designed and implemented, which includes data collection, web application, erotic image detecting and data store four sub-systems. The experimental and practical data all indicate that our system is suitable to deploy in the IDC room to help monitor and head off the erotic images.