Evaluating Deep Learning and Traditional Approaches Used in Source Camera Identification
The study compared three source camera identification methods—Photo Response Non-Uniformity (PRNU), JPEG artifact analysis, and Convolutional Neural Networks (CNNs)—using four smartphone devices. Results showed that the JPEG artifact method was the most accurate overall at 0.90, followed by PRNU at 0.71. The CNN approach performed poorly with an accuracy of 0.29, suggesting traditional feature-based techniques are superior for this task.