Application Research Of Image Spam Filtering Based On Wavelet And Support Vector Machine

With the proliferation of computers and the development of the Internet, thenumber of Internet users in China reached564million, which the vast majority ofpeople have e-mail, the use of e-mail has become an essential daily communicationtool. At the same time the emergence of spam,which brings trouble to the people’swork and life. Spam wastes valuable network resources and causes a great threat tothe security of the computer systems. Image spam has become the main form ofspam, which brings a greater harm, therefore it is very important how filtering imagespam.This article describes the background, characteristics and hazards of imagespam, summed up the spam filtering method and its advantages and disadvantages,and elaborate on the theoretical knowledge of the wavelet transform and supportvector machine. For the characteristics of image spam, In this paper, proposed asingle spam image recognition method, and designed of the filter frame andalgorithms.In the proposed filtration method, discrete haar wavelet color edge detection ofimage spam in the preprocessing stage, which contains the R, G, B three-color, andcontains more information than the gray level. Image spam are generally syntheticand contains a lot of text messages. Compared image spam which has morehorizontal and vertical texture, and color distribution at the same time is alsodifferent to the normal mail. Therefore, while using color information and edgefeatures, which can improve the filtering effect of image spam with the supportvector machine. In comparative experiments, filter to achieve the best results byselecting different Gaussian kernel parameter values; Describes the advantages anddisadvantages of different support vector machines by selecting a different kernelfunctions; Finally, comparison with other convenience advantages of the method isthat the accuracy rate.Experimental results show that image spam filtering method proposed in thispaper is feasible, and this method has a relatively good performance in error rate andprecision rate.