Two-Stream Neural Networks for Tampered Face Detection

TL;DR

This paper introduces a two-stream neural network approach combining global tampering artifacts and local noise residuals for improved face tampering detection, validated on a newly created dataset of tampered images.

Contribution

The paper presents a novel two-stream network architecture and a new dataset for face tampering detection, enhancing detection accuracy over existing methods.

Findings

The two-stream network outperforms single-stream models.

The dataset of 2010 tampered images enables robust evaluation.

Experimental results confirm the effectiveness of the proposed method.

Abstract

We propose a two-stream network for face tampering detection. We train GoogLeNet to detect tampering artifacts in a face classification stream, and train a patch based triplet network to leverage features capturing local noise residuals and camera characteristics as a second stream. In addition, we use two different online face swapping applications to create a new dataset that consists of 2010 tampered images, each of which contains a tampered face. We evaluate the proposed two-stream network on our newly collected dataset. Experimental results demonstrate the effectiveness of our method.