Image Spam Detecting Based On Combinatorial And Statistical Classifier

Along with the era of Internet of Things, email still is the most important and popular mean of communication. However, its byproduct----spam has brought a potential danger for people's lives. Image spam pushed the anti-spam technology into a new height. So how to detect image spam accurately and efficiently is an urgent problem.The thesis analyses the background, development actualities and significance of the studies on image spam systematically, studies the key technology of image spam deeply. The paper put forth the image spam detecting method based on the combinatorial and statistical classifier. The main work and contribution come out of the thesis are:(1) The SURF is used in order to extract the local invariant features. Then the Gaussian Mixture Model is designed to standardize the feature. As to cluster the datasets, the K-means algorithm is improved by means of using the Cross Entropy as the distance standard. Thus the Gaussian Mixture Model Classifier based on the Cross Entropy is designed, which has ideal results according to the simulation experiment.(2) Considering the spam image is always made of text and the image itself, the thesis put forths the stack combination model, which combines the Bayesian classifier to discriminate the spam based on the text and the SGMMKmeans classifier to discriminate the spam based on the local invariant features. The results of the simulation experiment show a high accuracy and recall.