<![CDATA[This study aims to develop and evaluate methods for digital image source device identification through three main approaches, namely EXIF metadata feature extraction, visual analysis using Convolutional Neural Networks (CNN), and Photo Response Non-Uniformity (PRNU). The dataset consists of 500 original images captured from five different devices, with 100 images per device containing intact metadata. The EXIF-only model was built using the Random Forest algorithm, the CNN model employed a ResNet18 architecture, while PRNU utilized high-pass filtering to construct sensor noise templates for each device. Evaluation was carried out using accuracy, precision, recall, and f1-score metrics. The results show that EXIF-only achieved perfect accuracy (100%) on the dataset with complete metadata, CNN reached 21% accuracy with imbalanced recall across classes, and PRNU demonstrated low performance due to the limited number of templates and image quality. These findings indicate that EXIF-only excels under intact metadata conditions but is vulnerable to manipulation, CNN can be applied when metadata is unavailable but requires optimization, while PRNU has potential resilience against metadata manipulation but demands higher-quality data. The novelty of this study lies in its comparative multi-method approach that integrates metadata-based, visual-based, and sensor fingerprint-based analyses, along with the proposal of a multimodal integration framework to enhance the reliability of device identification systems in digital forensic practice.]]>
Copyrights © 2025
