Research On Detection Method Of Phishing Based On CADW Integration

In recent years,there is a rapid growth of information fraud-related security incidents,on the one hand,the spread range of phishing websites have become more and more widely;one the other hand,the means of attack have become more and more diversified.So how to comprehensively and efficiently analyze the content of data is becoming more and more challenging,and has received more and more attention and exploited.From the nature of phishing websites,phishing attackers usually use certain pictures from legal websites to disguise.At the same time,in order to circumvent detection,they use more pictures for camouflage.Therefore,phishing detection based on image similarity becomes more and more important.SIFT performs well in image feature extraction because it can preserve the scale invariance and can fit the variability of phishing websites well.But under the influence of certain factors,if there are a large number of regions with similarity in shape in a picture,simply using the SIFT method for image matching will be some mismatches.To deal with it,we propose a new method,it maintain geometric consistency constraint to eliminates points where the slope of the neighborhood is not similar when dealing with the feature matching;at the same time,we add Pearson correlation coefficient constraints and further filter eigenvectors to effectively improves the problem that the original matching algorithm has small constraints and it does not consider the spatial geometric constraint information,and reduces the false matching rate.Performing contrast tests on images that have undergone noise,rotation scaling,lighting changes,and image viewing angle changes,the result shows that the improved matching algorithm can effectively reduce the mismatch rate.Compared with the original SIFT algorithm,the mismatch rate is reduced by 4 to 6 percentage points and the methods is more stable.More importantly,different phishing websites show different inducing messages.Single-level phishing detection is easily targeted by phishing attackers and detection results are difficult to meets the requirements of practical applications.To solve this problem,we proposed an effective method which is called CADW integration method,which calculates the correlation between classifiers and generates the weights of each classifier is automatically based on the correlation between the classifier and other basic classifiers.Weighted integration of the results of multiple related classifiers.The Experiment compares CADW integration methods with other commonly used integration methods.It uses legitimate websites and phishing website data sets crawled on authoritative websites.To construct six basic classifiers it uses three kinds of algorithms as the basic classifier:NBC,SVM and improved SIFT matching algorithm.The results show that the CADW integration method given in this paper has higher accuracy and recall rate.