Su Purui

DepSim: a dependency-based malware similarity comparison system

It is important for malware analysis that comparing unknown files to previously-known malicious samples to quickly characterize the type of behavior and generate signatures. Malware writers often use obfuscation, such as packing, junk-insertion and other means of techniques to thwart traditional similarity comparison methods. In this paper, we introduce DepSim, a novel technique for finding dependency similarities between malicious binary programs. DepSim constructs dependency graphs of control flow and data flow of the program by taint analysis, and then conducts similarity analysis using a new graph isomorphism technique. In order to promote the accuracy and antiinterference capability, we reduce redundant loops and remove junk actions at the dependency graph pre-processing phase, which can also greatly improve the performance of our comparison algorithm. We implemented a prototype of DepSim and evaluated it to malware in the wild. Our prototype system successfully identified some semantic similarities between malware and revealed their inner similarity in program logic and behavior. The results demonstrate that our technique is accurate.