As one of the topological properties of complex networks, the community structure has important theoretical and practical significance. We propose a community detection algorithm based on node dependence and the fusion of similar communities. The algorithm firstly divides the whole network into several local networks with large average clustering coefficients, thus constructing a skeleton of the complex networks. Then according to the definition of connectivity, the algorithm continuously absorbs the edge nodes of the community and small communities into the backbone network until all the nodes are accurately allocated to the community. This algorithm is applied to Zachary Karate Club network and the dolphin social network, and compared with the Girvan-Newman algorithm (GN) and Newman fast algorithm (NFA). The results show that our algorithm can effectively classify fuzzy edge nodes and the result of the community division has high accuracy.