Video tampering localisation using features learned from authentic content

Pamela Johnston,Eyad Elyan,Chrisina Jayne

doi:10.1007/s00521-019-04272-z

Pamela Johnston, Eyad Elyan + Show 1 more

Open Access

https://doi.org/10.1007/s00521-019-04272-z

Copy DOI

Abstract

Video tampering detection remains an open problem in the field of digital media forensics. As video manipulation techniques advance, it becomes easier for tamperers to create convincing forgeries that can fool human eyes. Deep learning methods have already shown great promise in discovering effective features from data, particularly in the image domain; however, they are exceptionally data hungry. Labelled datasets of varied, state-of-the-art, tampered video which are large enough to facilitate machine learning do not exist and, moreover, may never exist while the field of digital video manipulation is advancing at such an unprecedented pace. Therefore, it is vital to develop techniques which can be trained on authentic or synthesised video but used to localise the patterns of manipulation within tampered videos. In this paper, we developed a framework for tampering detection which derives features from authentic content and utilises them to localise key frames and tampered regions in three publicly available tampered video datasets. We used convolutional neural networks to estimate quantisation parameter, deblock setting and intra/inter mode of pixel patches from an H.264/AVC sequence. Extensive evaluation suggests that these features can be used to aid localisation of tampered regions within video.

Highlights

Automated video analysis is an increasingly important area of research
We show convolutional neural network (CNN) can be trained to estimate different compression parameters such as quantisation parameter, intra- or inter-frame type and deblocking filter setting for standalone sequence patches with reasonable accuracy
With video manipulation techniques currently increasing at an unprecedented rate, it is vital to develop features that can detect tampering irrespective of the original tampering method

Summary

Introduction

Automated video analysis is an increasingly important area of research. Video content creates a unique visual record, but not all aspects of video content are apparent to human eyes and this is of particular relevance in today’s age of fake news and falsified video. There is an increasing urgency to develop techniques to detect evidence of video processing even when it is invisible to human eyes. This raises the important question: How do we develop useful features for visual data when we might not be able to perceive such features using our own biological sensors? Deep learning provides a good tool kit for feature discovery from data; it is necessarily data hungry In fields such as video tampering, a large, labelled and sufficiently varied dataset which encompasses multiple examples from many recent techniques does not yet exist, [4] and its recent successor [5] show great promise. We must develop new techniques to exploit features common to many data examples

Objectives

Findings

Discussion

Conclusion