Textural Features Analysis Of Android Code

In recent years,with the rapid development of mobile Internet and Internet of things,the situation of network attack and defense is becoming increasingly severe.Malicious code,originally written by individuals,is now being transformed into teams and organizations that start large-scale and collaborative development.The detection of malicious code has developed from the original code detection to static and dynamic behavior analysis,even with artificial intelligence and big data technology.Facing the complex problem caused by current malicious' s characteristics of code dynamic mutation,polymorphism hidden,team aggregation and fusion of social work,a new method is used,plans to apply texture analysis of code to distinguish the behavior of code programmers behind the code,so as to provide support for the malicious code analysis and tracing.In order to study the calling relation between functions of Android code,we analyze and test the function call relation among different codes,so as to find the difference between different writing styles.The call relation between functions can be modeled into a directed graph,and the texture quantization of directed graphs formed by the relation of function calls is studied through three eigenvalues.The first eigenvalue is the level of value,the function calls should ideally be a tree,but in fact nodes of different layers will call each other,this caused a confusion between levels,the numbers of low level function calling high level and same level function is counted as level of value;the second eigenvalue is semi variance,image texture analysis usually uses semi variance to judge the image structure,the semi variance in horizontal direction of 1 degrees of the graph matrix is computed to judge the order of function;the last eigenvalue is the order value of distribution of node degree of the call graph,the random degree distribution of nodes is Poisson distribution,and the code written by the development team will be to more or less orderly,the difference between the degree distribution and the Poisson distribution of function calls is compute,called the ordered value of the degree distribution.These three eigenvalues was extracted from different samples and used to evaluate the difference between these three values for different styles of writing.The experimental results show that the degree of order distribution is the best,and the difference between the hierarchical and ordered values is not as good as the degree of order of degree distribution,which is suitable for distinguishing code with approaching number of functions.