LAUSR

Palmprint recognition is critical for high-security applications, such as access control and forensic investigations, due to its rich information content and resistance to forgery. However, extracting reliable features from low-quality palmprints, common in real-world scenarios like latent prints at crime scenes, remains challenging. Recent high-resolution palmprint research has focused on image preprocessing and feature extraction; however, errors introduced during preprocessing can compromise feature reliability, thereby degrading recognition accuracy. In this article, we propose an optical diffraction field-based method that extracts frequency-domain features from the grating-like ridge patterns of palmprints. Feature matching is evaluated using similarity measures including structural similarity index, Pearson correlation coefficient, and cosine similarity, with a random forest classifier for decision fusion. This method simplifies preprocessing, reduces computational complexity, and enhances robustness against noise and deformations. Experimental results on the THUPLMLAB dataset (a publicly available high-resolution palmprint database) achieve an equal error rate (EER) of 1.00%, demonstrating competitive performance against state-of-the-art methods reliant on intensive preprocessing. The proposed method provides a physically interpretable, efficient, and robust solution for biometric palmprint recognition.