As the use of electronic documents are becoming more popular, people want to find documents completely or partially duplicate. In this paper, we propose a near duplicate text detection framework using signatures to save space and query time. We also propose a novel signature selection algorithm which uses collection frequency of q-grams. We compare our algorithm with Winnowing, which is one of the state-of-the-art signature selection algorithms. We show that our algorithm acquires much better accuracy with less time and space cost. We perform extensive experiments to verify our conclusion.