The Design Of Networking Information Filtering System Based On Image Content

With the rapid development of computer networking and Internet, multimedia information, especially digital image information, is broadly used. However, the contents of the images in Internet are various, and lots of porngraphic images obstruct children's healthy growth. So, how to monitor and manage porngraphic images in Internet so that children can't contact them has become a challenging research subject. This paper deploys the research on techniques of Content-Based Image Filtering (CBIF). The paper points out their deficiencies after the briefly reviewing of traditional networking information filtering methods, and then designs a set of networking information filtering system based on a set of Content-Based Image Classification (CBIC) system, which has been implemented.The paper first introduces some related techniques to the research subject, such as image file format, skin color detection, skin texture detection, edge detection and shape descriptor, etc. The design and implementation of CBIC system is discussed in detail. The system mainly consists of skin color detector, skin texture detector, edge detector and shape descriptors extractor, feature extractor and classifier, totally 6 modules. When a new image comes, skin color detector and skin texture detector detect it in sequence. Then, the image is segmented according to the result of previous detections. After the segmentation, the system extracts the edge and shape descriptors of the image using edge detector and shape descriptors extractor. The feature vector composed of 11 features (in 4kinds), are extracted from the results of these detections and extractions. Finally, the classifier draws the conclusion built on the feature vector. During the procedure, we advance and improve some methods and algorithms. The first, refer to other CBIF systems, we bring out our CBIF system architecture. The second, during the skin color detection procedure, we don't employ the gray level statistics method directly, however, we adapt it to color level deviation method. The third, shape descriptors are used as a part of classification features. The last, during the feature extraction procedure, we draw a algorithm that detects the count of mazes without routine, refer to the maze routine algorithm. In addition, tow kinds of classifiers, Bayes classifier and SVM classifier, are implemented and their classification effects are compared. In the experiments of the CBIC system, the accuracy of classification is above 75%.Based on the CBIR system, the paper designs a set of networking information filtering system. The filtering system synthesizes the CBIF techniques and traditional filtering techniques, such as keyword filtering, packet filtering and URL blocking, so that it can filter text and image networking information fairly well. The filtering system is composed of 4 modules, i.e., data acquisition, data analysis, data processing and control interface. Because of the filtering system's adoption of Winsock2 SPI, it can capture and filter networking data packets conveniently.