Several graph summarization approaches aggregate dense sub-graphs into super-nodes leading to a compact summary of the input graph. The main issue for these approaches is how to achieve a high compression rate while retaining as much information as possible about the original graph structure within the summary. These approaches necessarily involve an algorithm to mine dense structures in the graph such as quasi-clique enumeration algorithms. In this paper, we focus on improving these mining algorithms for the specific task of graph summarization. We first introduce a new pre-processing technique to speed up this mining step. Then, we extend existing quasi-clique enumeration algorithms with this filtering technique and apply them to graph summarization.