Detecting malware variants via function-call graph similarity

Currently, signature-based malware scanning is still the dominant approach to identify malware samples in the wild due to its low false positive rate. However, this approach concentrates on programs' specific instructions, and lacks insight into high level semantics; it is enduring challenges from advanced code obfuscation techniques such as polymorphism and metamorphism. To overcome this shortcoming, this paper extracts a program's function-call graph as its signature. The paper presents a method to compute similarity between two binaries on basis of their function-call graph similarity. The proposed method relies on static analysis of a program, it first disassembles the program into assemble code, and then it uses a novel algorithm to construct the function-call graph from the assembly instructions. After that, it proposes a simple but effective graph matching method to compute similarity between two binaries. A prototype is implemented and evaluated on several well-known malware families and benign programs.