Shoeprint image retrieval and crime scene shoeprint image linking by using convolutional neural network and...

Area of Science:

• Forensic Science
• Computer Vision
• Pattern Recognition

Background:

• Shoeprint analysis is crucial in forensic investigations but challenging due to imperfect crime scene impressions.
• Existing methods struggle with partial, distorted, or faint shoeprints.

Purpose of the Study:

• To develop an automated shoeprint image retrieval system for forensic applications.
• To enhance the accuracy of matching crime scene shoeprints against reference databases.
• To extend the method for linking shoeprints between different crime scenes.

Main Methods:

• Utilized a pre-trained convolutional neural network (CNN) for feature extraction from pre-processed shoeprint images.
• Employed normalized cross-correlation (NCC) to calculate similarity scores between extracted features.
• Applied the method to both image retrieval and linking of shoeprints from distinct crime scenes.

Main Results:

• Achieved 82% retrieval accuracy, defined as the correct shoeprint appearing in the top 1% of results.
• Demonstrated 88.99% accuracy in linking shoeprints from different crime scenes when the match was within the top 20% of results.

Conclusions:

• The proposed CNN and NCC-based method significantly improves shoeprint retrieval accuracy in forensic analysis.
• The system's ability to link crime scenes based on shoeprints offers a valuable tool for investigators.
• This approach addresses the challenges posed by imperfect shoeprint evidence effectively.