Feature Point Analysis For Image Spam E-mail Detection

With the popularity of the Internet, e-mail, which is convenient, fast and low-cost, has become one of the main communication tools on the Internet. Meanwhile, spam is becoming more and more harmful. Recently, Image-based spam is becoming a new threat to the Internet and its users. Therefore, how to deal with image-based spam becomes an emergency task for Internet security.In the early work of our research group, we proposed an image filtering system which detects the spam image by matching it with user-specified image content. The experiments showed that there are two advantages in that system. The rules are varying from person to person and the system is easy to implement. However, this system also has two disadvantages. The first one is that its matching algorithm is not efficient enough. The other one is that the user-specified images maybe affect the efficiency of the system.In this paper, we propose solutions to solve the above problems. First, we present a novel image matching algorithm. In this algorithm we use DoG to extract image features, and then adopt geometry transform to judge whether two images are matching. This algorithm has improved the performance of the system effectively. To solve the second problem, we use Mean Shift algorithm to cluster the feature points and then locate the highest density area of the feature points, which is helpful for reducing blindness of users' operation.The experimental results have demonstrated the feasibility and validity of our proposed methods.