Work

On Combining Deep Neural Network Classifiers for Source Device Identification

Year: 2025

Type: article

Abstract: This paper proposes combining deep neural network classifiers while simultaneously optimizing the networks. The proposed combination scheme enhances the accuracy of each classifier, which, in turn, boosts the overall combined accuracy during a post-processing step. The proposed classification scheme is thoroughly evaluated on a dataset specifically designed for multimedia forensics research. The combined classifiers include shallow and deep neural networks, with input data comprising original and manipulated content processed through online social networks such as YouTube, WhatsApp, and Facebook. The experimental results demonstrate promising performance, proving the usability of the proposed classifier combination scheme. Specifically, it is observed that the accuracy of shallow neural networks improves significantly when combined with deep neural networks. This performance enhancement is particularly notable when the combined classifiers are trained on data manipulated by online social network platforms.

Source: IEEE Access

DOI: 10.1109/access.2025.3555141

Authors: Ioannis Tsingalis, Constantine Kotropoulos

Institution: Aristotle University of Thessaloniki

Language: English

Topic: Digital Media Forensic Detection

Subfield: Computer Vision and Pattern Recognition

Field: Computer Science

Domain: Physical Sciences

Open Access status: gold

Estimated APC paid: $2,075

Funder: Hellenic Foundation for Research and Innovation