Abstract: Since Android applications（abbr. APPs） are easily cracked and repackaged, there are thousands of pirated APPs on the Internet. The problem that how to calculate the similarity among APPs quickly and accurately becomes one of the hotspots of Android software protection. In order to improve the efficiency of similarity detection, a novel framework that based on property classifying is proposed in this paper. Firstly, the framework tries to divide a collection of APP into two different sub-collections according to several properties of APP. Secondly, features and similarity will be extracted and calculated parallel based on properties of every APPs in sub-collections. Finally, the result of similarity detection for the APP collection will be gathered from two sub-collections. Experimental result shows that the proposed framework can cut down about 50% detection time while maintaining a similar accuracy.

Key words: Android, software similarity, property classifying, native code

摘要： 由于Android应用（Android Applications，APP）容易被逆向修改而产生大量的盗版，因此如何快速准确地判断APP之间的相似性是目前Android软件保护的研究热点之一。从提高APP相似性检测效率的角度出发，提出了一种基于属性分类的APP相似性检测框架。该框架首先利用APP的某些属性对需要检测的APP集合进行分类，将差异较大的APP划分到两个不同的子集，然后采用并行的方式分别对两个子集进行特征提取和相似性计算。实验结果表明，该框架采用的事先分类和并行计算的方法使得相似性检测的效率得到了明显的提高，在检测正确率相近的情况下可以缩短近一半的检测时间。

关键词: Android, 软件相似性, 属性分类, 本地代码