PalmNet: A Robust Palmprint Minutiae Extraction Network

Minutiae is the crucial foundation of palmprint recognition due to their discriminant and persistence. However, disturbed by complex latent noise, wide creases, and various frequency changes, it still challenging to extract palmprint minutiae. To address these issues, a robust palmprint minutiae extraction framework, PalmNet, is proposed. Specifically, considering the complex noise on the latent palmprint that is difficult to model, we avoid directly extracting the minutiae on the original image, but based on the Gabor enhanced map. Further, to make up for amplitude information missing in the Gabor enhanced phase field, we fuse phase and amplitude features to boost model performance. The orientation field is an inherent property of palmprints. Accurate orientation field estimation can benefit minutiae extraction and Gabor enhancement. Therefore, to recover the orientation field disturbed by wide crease, we make use of deep feature expression power and orientation consistency strategy to estimate orientation field. This two-stage method combines palmprint prior knowledge and has better interpretability, which can help us understand and analyze the total framework. Experiments on public LPIDB v1.0 and THUPALMLAB palmprint database demonstrate that our proposed algorithm outperforms the state-of-the-art. Notably, our algorithm improves the rank-1 recognition rate from 91.7% to 99.6% on the THUPALMLAB database.