Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

From Images to Detection: Machine Learning for Blood Pattern Classification

Yilin Li, Weining Shen

TL;DR

This paper tackles the problem of distinguishing gunshot backspatter from impact spatter in bloodstain pattern analysis by building an interpretable feature pipeline from ellipse-based stain extraction, followed by region-based summarization to create pattern-level descriptors. It introduces the Stability Importance Score (SIS) to robustly gauge feature importance across multiple model fits and demonstrates that XGBoost, complemented by region-aware and shading features, achieves high classification accuracy (up to ~$92$–$97 ext{%}$ depending on metric) on a public dataset. The work also compares with Random Forest and other classifiers, showing superior performance for ensemble methods and highlighting data quality as a key factor for cross-dataset generalization, with limited transfer to other datasets. Overall, the approach provides a quantitative, scalable framework for BPA decision-support and sets directions for richer feature spaces and more diverse datasets.

Abstract

Bloodstain Pattern Analysis (BPA) helps us understand how bloodstains form, with a focus on their size, shape, and distribution. This aids in crime scene reconstruction and provides insight into victim positions and crime investigation. One challenge in BPA is distinguishing between different types of bloodstains, such as those from firearms, impacts, or other mechanisms. Our study focuses on differentiating impact spatter bloodstain patterns from gunshot bloodstain patterns. We distinguish patterns by extracting well-designed individual stain features, applying effective data consolidation methods, and selecting boosting classifiers. As a result, we have developed a model that excels in both accuracy and efficiency. In addition, we use outside data sources from previous studies to discuss the challenges and future directions for BPA.

From Images to Detection: Machine Learning for Blood Pattern Classification

TL;DR

This paper tackles the problem of distinguishing gunshot backspatter from impact spatter in bloodstain pattern analysis by building an interpretable feature pipeline from ellipse-based stain extraction, followed by region-based summarization to create pattern-level descriptors. It introduces the Stability Importance Score (SIS) to robustly gauge feature importance across multiple model fits and demonstrates that XGBoost, complemented by region-aware and shading features, achieves high classification accuracy (up to ~– depending on metric) on a public dataset. The work also compares with Random Forest and other classifiers, showing superior performance for ensemble methods and highlighting data quality as a key factor for cross-dataset generalization, with limited transfer to other datasets. Overall, the approach provides a quantitative, scalable framework for BPA decision-support and sets directions for richer feature spaces and more diverse datasets.

Abstract

Bloodstain Pattern Analysis (BPA) helps us understand how bloodstains form, with a focus on their size, shape, and distribution. This aids in crime scene reconstruction and provides insight into victim positions and crime investigation. One challenge in BPA is distinguishing between different types of bloodstains, such as those from firearms, impacts, or other mechanisms. Our study focuses on differentiating impact spatter bloodstain patterns from gunshot bloodstain patterns. We distinguish patterns by extracting well-designed individual stain features, applying effective data consolidation methods, and selecting boosting classifiers. As a result, we have developed a model that excels in both accuracy and efficiency. In addition, we use outside data sources from previous studies to discuss the challenges and future directions for BPA.
Paper Structure (14 sections, 9 figures, 2 tables)

This paper contains 14 sections, 9 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Data
  3. Data Source
  4. Image Processing
  5. Feature Engineering
  6. Stain Features
  7. Summarized Features for Patterns
  8. Statistical Analysis
  9. Exploratory Data Analysis
  10. Models and Results
  11. XGBoost
  12. Random Forest
  13. Other Classifiers
  14. Discussion

Figures (9)

  • Figure 1: On the right is a sample of impact spatter pattern; the scale is on the edge. The test rigs used for generating the patterns is shown on the left. The figures are taken from attinger2018data.
  • Figure 2: A gun shot pattern sample. The figures are taken from attinger2019data.
  • Figure 3: An image undergoing processing steps.
  • Figure 4: Example image undergoing processing and identification steps.
  • Figure 5: Boxplot of Features Group by Label gun, for feature mean_maj_len, mean_min_len, mean_area, mean_solidity, and fract1_ring_15_25, where gun = 1 stands for gun-shot patterns.
  • ...and 4 more figures