The Detection And Filtering Of Mobile Internet Spam Image

With the rapid development of mobile Internet,it is convenient for people to get the information,however meanwhile a large amount of spam has affected people's work and life.Spam is usually transmitted in the form of sounds,texts,videos,and images.The spam of advertisements and pornography which cause bad social influence are mainly transmitted in the form of images or pictures.Therefore,it is imperative to study the accurate and efficient detection and filtering methods of spam images.According to the characteristics of spam images in the environment of mobile Internet,starting from the image content,by analyzing and summarizing the content and characteristics of spam images,a classification method based on SVM is designed and conducted in this study.Firstly,the image sets including varieties of images are divided into training images and test images,whose sizes are unified with MATLAB's imresize function.Then the weighted average method is used to deal with the images,and the HOG feature and LBP feature of the training images are extracted.The generated eigenvectors are used to train the SVM classifier,and then the test images are put into the trained classifier for classification.Through hundreds of experiments,The SVM classification filtering ability is low whether it is based on HOG features or LBP features.This shows,using single feature to represent the image can reduce the classification accuracy of classifier.This problem can be solved by designing and implementing a sparse representation classification method based on HOG and LBP.After the training images are preprocessed,HOG feature and LBP feature are extracted.The two eigenvectors are generated by serial fusion to form the final fusion eigenvector.Then the test images are represented as a linear combination of the atoms in all the training images through the spare representation,and the convex relaxation is used to solve the sparse representation coefficients.Finally,the reconstruction error is obtained by using the weighted linear combination of all training images' sparse coefficients.The experiment results show that the sparse representation algorithm has stronger robust and dynamic characteristic than SVM,and it is better at improving adaptability and classification filtering capability.