Research On Content-Based Image Spam Filter Technology

Many content-based spam filtering techniques which combine the development of machine learning, text categorization and information filtering have been carried out and widely used in recent years, but these means have certain limitations. Because these technologies are incapable of filtering image-based spam, and with the more and more increment of the image-based spam, so how to identify and filter it is becoming a very important problem that need IT realm and the mail server provider to be resolve urgently.In this paper, Analyzed the research condition of spam filter at the present time, which mainly include the definition and endanger of the spam, the current dominating spam filtering techniques and its merit and demerit. Image feature abstraction is analyzed as a key problem of image-based spam filtering. The various visible features and feature extraction method have been studied systematically, the contents includes color, texture, shape, etc.To contrapose of the characteristic of the image-based spam, this paper proposed a new kind of image spam filter method based on similarity detection of image, which in terms of the spam sending behavior and the content that includes send in bulk, repeatedly and highly resemble content. the method implement based on similarity detection between new mail image and the sample-image of spam: extract the low-level visual features of image, which include color feature, texture feature and shape feature, then judging the new mail image is a spam image or not by detect similarity between new mail image and the sample-image of spam according to the combined vision features. At the same time, the several related problem and key techniques have been discussed, which include similarity measuremet and feature normalization, etc.It is showed with experiments that this new method which based on similarity detection of image has a good performance. It does some useful exploring for the way of image-based spam filtering, and may provide solid theoretical support for designing the anti-spam project; its research has both the theory and the application value.