Method for filtering image-based junk mails

Abstract

The invention discloses a method for filtering image-based junk mails, which includes the steps of collecting a sample of the image-based junk mails to establish a training set of the image-based junk mails; extracting Fourier-Mellin invariant features of training set images from the training set as low level features of the images; carrying out analysis and dimensionality reduction to the Fourier-Mellin invariant features to obtain a low-dimensional feature vector for classification; and adopting the low-dimensional feature vector for training a classifier model and adopting the classifier model for classifying samples of new mails. The adoption of the method greatly reduces hazards caused by harassment of the junk mails for normal e-mail users.

Description

technical field [0001] The invention relates to the field of computer application technology, in particular to a method for filtering image spam. Background technique [0002] While the email system provides people with convenient means of communication, it also leads to the flooding of junk information on the Internet. The content of spam is full of commercial fraud, violence, pornography, reactionary political propaganda, etc., and it has gradually become the main carrier of computer viruses, seriously affecting the healthy development of the Internet. [0003] With the continuous development and improvement of spam detection and filtering technology, many new forms of spam are constantly derived. Among them, image spam is an effective way to evade the currently widely used filtering technology for email text content. Image spam refers to the mail form in which spam creators embed spam information into images, and then use various image transformation methods to further ...

AI Technical Summary

Problems solved by technology

However, optical character recognition can only be used when the image background is relatively simple, and the character deformation is not particularly serious, and there are disadvantages such as high algorithm time complexity and large limitations.

Images