The enhancement of fingerprint plays an important role in automatic fingerprint identification system. In order to make up for the shortcomings of the traditional fingerprint enhancement, this study proposes a novel algorithm by using orientation Gaussian bandpass filter (OGBPF) to enhance the fingerprint firstly, and then the deep Boltzmann machine (DBM) with orientation selection is employed to reconstruct these regions that are enhanced incorrectly in the first phase. The fingerprint is enhanced based on the quality grading scheme and the composite window strategy. In the proposed method, the traditional enhancement method and deep learning method complement one another perfectly. To validate the performance, the proposed method has been applied to fingerprint enhancement on the FVC2004 databases. Experiments show that, compared with the state-of-the-art enhancement methods, the proposed method is more accurate and more robust against noise, and can achieve better results.