Design A Fast Filtering Systerm For Image Spam And Research On Approximate Matching Alogrithms

In recent years, image spam has been exploding so rapidly to become the major form of spam. However, the existing spam filtering system lacks the ability of analyzing image spam. Image spam filtering technology became an important topic of spam filtering.This paper designed a fast filtering system for image spam and implemented the structure of it, which is designed to improve existing spam filtering system’s shortages and provide a research platform for further key technology of image-based spam filtering problem. The system’s framework has been built, which can achieve bulk mail file parsing, image based spam image data sets and approximate matching Image spam. Firstly, this paper introduce Image spam filtering technology research and analyze the needs of the system to complete the system framework for the design and the division of the system modules. Secondly, this paper introduce the process of system software interface’s designing and building, system data table’s designing and building. Finally, elaborated mail parsing module and approximate matching module based on image characteristics of the system implementation process. Finally, this paper detailed description of the building process of system’s mail analytic module and system’s approximate matching module.Besides what mentioned above, this paper also complete the image local feature’s matching test through the system’s framework, which based on SIFT, SURF and ORB algorithm. The results showed that the SIFT algorithm is not sensitive to a variety of changes in an image with a stable matching accuracy, but its feature matching process is time-consuming. This paper attempts to improve the feature matching process of SIFT algorithm through combining the SIFT algorithm and BOW algorithm. BOW algorithm combined with SIFT algorithm can effectively improve the feature matching speed and high matching accuracy can still ensure with a reasonable threshold.