A Multimodal Fusion Approach for Bullet Identification Systems.

Area of Science:

• Forensic Science
• Ballistics
• Image Analysis

Background:

• Firearm identification relies on unique microscopic striations left on bullets by gun barrels.
• Current bullet identification methods are broadly categorized into 2-D and 3-D based on image acquisition.
• 2-D optical imaging is a common technique for analyzing bullet striations.

Purpose of the Study:

• To develop and evaluate a novel multimodal 2-D bullet identification technique.
• To enhance the accuracy and reliability of firearm identification using optical images.
• To optimize the fusion of data from different imaging modalities.

Main Methods:

• A multimodal approach using 2-D optical images was developed.
• A rule-based linear weighted fusion method combined semantic-level decisions from various modalities.
• A genetic algorithm was employed to optimize the weights for modality fusion.

Main Results:

• The proposed multimodal approach was tested on a dataset of 180 2-D bullet images from 90 AK-47 barrels.
• The method demonstrated superior performance in bullet identification compared to conventional techniques.
• Optimized fusion weights significantly contributed to improved identification accuracy.

Conclusions:

• The multimodal 2-D bullet identification method offers a promising advancement in forensic science.
• The genetic algorithm-optimized fusion approach enhances the effectiveness of combining different imaging modalities.
• This technique provides a more accurate and reliable tool for firearm identification.