Skin injury model classification based on shape vector analysis.

Area of Science:

• Forensic Science
• Biometrics
• Computer Vision

Background:

• Judicial decision-making relies on forensic expert classification of skin injuries, which is often subjective.
• This study investigates objective classification of simulated skin injuries using 3D surface models and numerical shape descriptors.

Purpose of the Study:

• To determine if known classes of simulated skin injuries can be statistically classified using 3D surface models.
• To evaluate the accuracy of automated classification based on derived numerical shape descriptors.

Main Methods:

• Simulated skin injuries (abrasions, incised wounds, gunshot wounds, strangulation marks, patterned injuries) were created using plasticine.
• 3D surface scanning captured models, and numerical shape descriptors (e.g., mean curvature, convex hulls, Fourier transforms) were derived.
• Regularized Discriminant Analysis (RDA) was optimized and applied for classification after dimensionality reduction.

Main Results:

• Receiver Operating Characteristic analysis yielded an ideal Area Under the Curve of 1.0 for all six injury categories.
• Predictive RDA achieved an average correct recognition rate (CRR) of 97.22% using k-fold cross-validation.
• Adding uniform noise degraded the average CRR to 71.3%, highlighting the importance of surface quality.

Conclusions:

• Automated classification of simulated skin injuries is feasible using digitized 3D surface shape data and derived descriptors.
• The method provides an objective basis for discriminating between injury categories in medicolegal analysis.
• Technical surface quality is critical for accurate classification, as noise significantly impacts performance.