Latent fingerprints are important for identifying criminal suspects. However,
recognizing a latent fingerprint in a collection of reference fingerprints
remains a challenge. Most, if not all, of existing methods would extract
representation features of each fingerprint independently and then compare the
similarity of these representation features for recognition in a different
process. Without the supervision of similarity for the feature extraction
process, the extracted representation features are hard to optimally reflect
the similarity of the two compared fingerprints which is the base for matching
decision making. In this paper, we propose a new scheme that can model the
pair-relationship of two fingerprints directly as the similarity feature for
recognition. The pair-relationship is modeled by a hybrid deep network which
can handle the difficulties of random sizes and corrupted areas of latent
fingerprints. Experimental results on two databases show that the proposed
method outperforms the state of the art.

研究使用混合深度网络建立两个指纹间相似性特征直接模型的方法，得到的结果在两个数据集上均优于现有技术。