Brand identification of roller-type date stamps based on impression features.

Area of Science:

• Forensic Science
• Document Examination
• Machine Learning Applications

Background:

• Roller-type date stamps are widely used in Chinese manufacturing and administrative sectors.
• Stamp impressions are crucial in forensic casework for forgery, counterfeit labeling, and traceability disputes.

Purpose of the Study:

• To develop a quantitative framework for brand identification of Chinese roller-type date stamps.
• To leverage morphological feature extraction and machine learning for stamp analysis.

Main Methods:

• Evaluated 16 commercial stamp models under 36 conditions, generating 17,280 impression samples.
• Extracted 17 morphological features from Chinese characters for 'year', 'month', and 'day'.
• Utilized Interval Overlap Rate (IOR) and Principal Component Analysis (PCA) for feature refinement and employed four supervised classifiers (GBDT, RF, SVM, NB).

Main Results:

• Gradient Boosting Decision Tree (GBDT) and Random Forest (RF) achieved high accuracies (96.76% and 95.66% in lab; 89.06% and 90.88% in field).
• Demonstrated that stamp impressions possess quantifiable, brand-specific morphological characteristics.
• Validated the robustness and discriminative ability of the data-driven models.

Conclusions:

• Roller-type stamp impressions can be reliably identified using data-driven machine learning models.
• The proposed approach provides a rapid, non-destructive, and scientifically robust tool for forensic brand identification.
• This method enhances evidential authentication in cases involving counterfeit labeling and product traceability.