UC Irvine

Shoeprints are a common type of evidence found at crime scenes and are regularly used in forensic investigations. However, their utility is limited by the lack of reference footwear databases that cover the large and growing number of distinct shoe models. Additionally, existing methods for matching crime-scene shoeprints to reference databases cannot effectively employ deep learning techniques due to a lack of training data. Moreover, these methods typically rely on comparing crime-scene shoeprints with clean reference prints instead of more detailed tread depth maps. To address these challenges, we break down the problem into two parts. First, we leverage shoe tread images sourced from online retailers to predict their corresponding depth maps, which are then thresholded to generate prints, thus constructing a comprehensive reference database. Next, we use a section of this database to train a retrieval network that matches query crime-scene shoeprints to tread depth maps. Extensive experimentation across multiple datasets demonstrates the state-of-the-art performance achieved by both the database creation and retrieval steps, validating the effectiveness of our proposed methodology.