Matching unknown latent fingerprints lifted from various objects or surfaces at crime scenes to fingerprints of known subjects is of vital importance for law enforcement agencies to identify suspects. Banknotes are one of the most common objects containing valuable latent fingerprints. However, due to the complex pattern printed on banknotes, it is a challenging problem even for human experts to mark minutiae in such fingerprints. In this paper a novel technique is proposed to enhance fingerprints on banknotes so that they can be successfully identified by existing fingerprint matchers. The proposed algorithm is based on subtraction of the reference orientation in reference banknote, which is registered to the latent fingerprint by a coarse-to-fine registration algorithm. Promising results are reported on a database which contain 192 latents on banknotes, which proves the effectiveness of the proposed algorithm.