A software system, called SHOECALC, is being developed at NIST. It is designed to help both researchers and footwear examiners in the assessment of metrics or scoring procedures that provide objective characterizations of correspondences and discrepancies between features from two footwear impressions being compared. The system is composed of five modules. Although these modules are currently only in skeleton form, they will eventually function as follows:

  1. SHOEBASE – Will be a database consisting of crime scene impressions and metadata; a catalogue of outsole designs and metadata; test impressions from shoes of arrestees; a catalogue of Randomly Acquired Characteristics (RACs) along with shape, size, location, brand, outsole design, etc.; and interfaces and formats for submitting and maintaining footwear data.
  2. SHOEGULI – Will be used for research and testing purposes, and will generate synthetic footwear impressions with ground truth known, and with user specified characteristics. These characteristics will include particular subpopulation of outsole designs, wear amounts, sizes, and distributions of RACs; also different matrix/substrate combinations. SHOEGULI will generate both synthetic test impressions and crime scene impressions (e.g., partial impressions).
  3. SHOEMET – Will be a workbench for experimentation with different similarity measures. Some measures lead to better discrimination between mated and non-mated pairs of images than others. The user will input a function for computing a similarity score and apply it to any given pair of images; a numerical score will be reported. SHOEMET will use SHOEGULI to conduct experiments and produce Receiver Operating Characteristic (ROC) plots for comparing with a catalog of known, high performance similarity measures.
  4. SHOEQ – Will provide quality measures - different characteristics that describe the degradation, distortion, completeness, and quantity of features in an impression. The user will input a footwear image and the output will be a list of quality scores. SHOEQ will be used to calculate better scores for casework, and as a workbench for researchers to develop better image quality metrics.
  5. SHOESHINY – Will be a GUI for user interaction with the other modules of SHOECALC. SHOESHINY will allow the user to upload images for calculation of similarity and quality scores; will allow the user to examine various choices of similarity metrics and their ROC curves, and select choices for reporting the information in the evidence; and will allow exploratory pattern analysis.

Various pieces of these modules have been implemented, including initial data collection for SHOECALC, generating soft synthetic (crime-scene-like) footwear data from actual test impressions for SHOEGULI, development of initial matching algorithms and similarity measures for SHOEMET, implementation of initial set of quality metrics for SHOEQ, and development of a preliminary user interface for SHOESHINY to view, compare and analyze patterns and dissimilarity scores for pairs of pattern

Martin Herman, Hari Iyer, Yooyoung Lee, Steve Lund, Gunay Dogan, Simone Gittelson
National Institute of Standards and Technology (NIST)