Image Spam Filtering Based On Content Analysis Research

In the ever-changing age of the Internet, e-mail has became an important component of people's work, study and life because of its convenient, fast, low-cost advantages. But the spam problem which is accompanied by e-mail is getting worse. Early spam filters mostly based on the text filtering methods, such as keyword filtering, Bayesian filtering, artificial intelligence and machine learning the language, and so on. These methods achieved high accuracy in spam filtering. However, the emergence of the image spam, significantly reducing the text-based spam filters'effectiveness. With the explosive growth of image spam, how to identify and filter image spam has became an urgent problem to resolve.To contrapose of the above problem, this thesis did an exploratory research on the filtering technology of image spam, which based on the analysis of image content. The main results are as follows:First, an overview of the current situation for spam filters, including the market forecasts of anti-spam, the development of spam filtering technology, the status of spam and anti-spam.Second, this thesis had a detailed analysis of image spam definition, development and construction, describes the types of image spam filtering methods.Third, discussion of the key issue of image spam filtering, and the feature extraction methods of image on color, texture and shape.Fourth, proposed an image spam filtering method based on similarity detection of image: extract the image feature on color, texture and shape, then judging the new mail image is a spam image or not by the measure of similarity between new mail image and the spam image sample.