Challenges in Modeling, Interpreting, and Drawing Conclusions from Images as Forensic Evidence

Abstract

When a crime is committed, law enforcement directs crime scene experts to obtain evidence that may be pertinent to identifying the perpetrator(s). Much of this evidence comes in the form of images, either digitally transcribed (e.g.,: fingerprints, handwriting), or as digital photographs (e.g., biometric images, photographs of patterns created by blood spatter or arson). Finding models that faithfully capture the “key features” in these images is critical: attribution of the evidence will be accurate only if these “key features” can be properly compared across different images. The huge variety in the types, shapes, and locations of such features leads to challenges in obtaining valid inferences. We describe some of these challenges, discuss some prior approaches, and suggest future directions which need to be pursued to avoid miscarriages of justice that have occurred in the absence of statistically-validated methods of inference for forensic evidence.