Overlapping community structure has attracted much interest in recent years since Palla et al. proposed the k-clique percolation algorithm for community detection and pointed out that the overlapping community structure is more reasonable to capture the topology of networks. Despite many efforts to detect overlapping communities, the overlapping community problem is still a great challenge in complex networks. Here we introduce an approach to identify overlapping community structure based on an efficient partition algorithm. In our method, communities are formed by merging peripheral clusters with cores. Therefore, communities are allowed to overlap. We show experimental studies on synthetic networks and real networks to demonstrate that our method has excellent performances in community detection.