Hierarchical Image Spam Filtering Technology Research

With the increasing Internet bandwidth and the improving detection rate of spam texts,the era of image spam has come.To be able to cope with the massive image spams,this paper build a comprehensive image spam filtering engine base on Postfix as a mail server framework.It is proposed by the following three techniques for image spam detection and recognition.Firstly,using the global image features,classify images quickly to filter.The color in the spam image is more monotonous and the characteristics of the text is more than normal.In addition,in order to avoid filtration,a large number of image spam are added to the image noise and other disturbances,thus the presence of interfering signals is also an important feature that can be used to determine the image spam.This topic will focus on the use of color,texture,the proportion of characters,image size and so on to estimate the spam image which can be effective when analyzed the spam image with strong disturbance signal.This method can not only improve the shallow analysis phase of image spam filter quality and efficiency,-but also improve the effect when analyzing the deep content on the image.Secondly,make use of the characteristics of local feature descriptors to build inverted index and achieve an approximate image retrieval filter.Using this method,not only can keep the local description characterization of the advantages of good anti-interference,eliminating the mismatch phenomenon,but also guaranteehigh computational efficiency and achieve massive real-time image processing.Finally,based on the email image text filtering technology,we according to the characteristics of the text in the spam images,proposethe reconstruction technology of character recognition result based on WAF model.Through the analysis of the surrounding text,we identify the key word for associating,produce the most likely keywords in this environment,and then through the analysis of the keywords,effectively discriminant image spam.This paper investigate the above three methods for experiments,and achieve good effect.