Authentication of forged inked fingerprints utilizing silicone molds.

Although the use of forged inked fingerprints is not common in criminal cases, it is gradually increasing in civil cases. This study introduces a rapid and nondestructive method for detecting forged inked fingerprints using Raman spectral, morphology, and deep learning. To develop an effective method to detect forged inked fingerprints, thereby enhancing the reliability of forensic evidence in judicial settings. The study explored Raman spectroscopy for differentiating genuine from forged inked fingerprints. The signals were examined by similarity and Hotelling T tests. Morphological analysis was conducted on 3600 inked fingerprints, focusing on external contours, ridge widths, and ridge discontinuity. Chi-square and Mann-Whitney U tests were used to evaluate the effectiveness of these features. A deep learning model (ResNet50_AuI) was developed by integrating Feature Pyramid Network (FPN) and multi-head self-attention (MHSA) into the residual network (ResNet). This model was trained and tested using a custom database. Raman spectroscopy alone could not distinguish between genuine and forged fingerprints. Morphological analysis showed that external contours were most useful for authentication, followed by ridge discontinuity. The ResNet50_AuI model achieved 98.88% accuracy, emphasizing the importance of external contours. This study evaluates three methods for authenticating inked fingerprints, highlighting the potential and limitations of each method in improving the integrity of forensic evidence.