New computer viruses are continually being generated and they cause damage all over the world. In general, current anti-virus software detects viruses by matching a pattern based on the signature; thus, unknown viruses without any signature cannot be detected. Although there are some static analysis technologies that do not depend on signatures, virus writers often use code obfuscation techniques, which make it difficult to execute a code analysis. As is generally known, unknown viruses and known viruses share a common feature. In this paper we propose a new static analysis technology that can circumvent code obfuscation to extract the common feature and detect unknown viruses based on similarity. The results of evaluation experiments demonstrated that this technique is able to detect unknown viruses without false positives.

Recently, the appearance frequency of computer virus variants has increased. Updates to virus information using the normal pattern matching method are increasingly unable to keep up with the speed at which viruses occur, since it takes time to extract the characteristic patterns for each virus. Therefore, a rapid, automatic virus detection algorithm using static code analysis is necessary. However, recent computer viruses are almost always compressed and obfuscated. It is difficult to determine the characteristics of the binary code from the obfuscated computer viruses. Therefore, this paper proposes a method that unpacks compressed computer viruses automatically independent of the compression format. The proposed method unpacks the common compression formats accurately 80% of the time, while unknown compression formats can also be unpacked. The proposed method is effective against unknown viruses by combining it with the existing known virus detection system like Paul Graham's Bayesian Virus Filter etc.