Rank Fusion-Based Crime Scene Shoeprint Image Retrieval

The primary objective of shoeprint retrieval is to identify shoeprints with comparable patterns among the numerous shoeprints collected from different crime scenes. This task becomes challenging when trying to find the most similar shoeprints from a large number of deteriorated prints. Most existing shoeprint retrieval methods focus on using either holistic or local features for matching, which may lead to reduced accuracy, especially with degraded shoeprints obtained from crime scenes. In this study, we propose a framework for retrieving crime scene shoeprint images by fusing handcrafted and deep features. The proposed approach allows us to prioritize candidate shoeprints by assigning them top ranks. Experiments are conducted on the publicly accessible FID-300 dataset, which consists of crime scene shoeprint images. The results demonstrate that the proposed approach can effectively deal with real crime scene shoeprint images, achieving a cumulative match score of nearly 98% within the top 20 percent.