Ever since social networks became the focus of a great number of researches, the privacy risks of published network data have also raised considerable concerns. To evaluate users' privacy risks, researchers have developed methods to de-anonymize graphs and identify same person in different graphs, yet the existing algorithms either requires high-quality seed mappings, or have low accuracy and high expense. In this paper, an effective and efficient seedless de-anonymization algorithm, "RoleMatch" is proposed. This algorithm is based on the network topology and consists of (1) a new cross-graph node similarity measurement "RoleSim++" with fast computation method, and (2) an effective node matching algorithm considering both similarities and feedbacks. In experiments, the algorithm is tested with graphs anonymized in several popular anonymization ways, using the data from LiveJournal. In addition to the traditional symmetric experiments, an asymmetric experiment setting is proposed to mimic closer to real-world application. The results from those experiment show that with the proposed algorithm the de-anonymization work achieves superior performance compared with existing de-anonymization algorithms.