Demosaicing to Detect Demosaicing and Image Forgeries

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

The trustfulness of images is a critical concern. Digital photographs can no longer be assumed to be truthful; indeed, digital image editing tools can easily and convincingly alter the semantic content of an image. Being able to analyse an image to check for forgeries becomes of the utmost importance in many domains, from police investigations to fact-checking and journalism. We propose here to analyse traces left by the camera during demosaicing, one of the first steps of image formation. When an object is added or displaced on an image, the demosaicing traces can be disrupted, leaving forgery clues. In order to detect these inconsistencies, we explore the possibilities offered by double demosaicing. Computing the demosaicing residual of an image with different demosaicing algorithms and patterns enables one to find image regions with inconsistent demosaicing traces. We render the method fully automatic by a simple a contrario scheme computing forgery detection thresholds with statistical guarantees on the number of false alarms.

Original languageEnglish
Title of host publication2022 IEEE International Workshop on Information Forensics and Security, WIFS 2022
PublisherInstitute of Electrical and Electronics Engineers Inc.
ISBN (Electronic)9798350309676
DOIs
Publication statusPublished - 1 Jan 2022
Externally publishedYes
Event2022 IEEE International Workshop on Information Forensics and Security, WIFS 2022 - Shanghai, China
Duration: 12 Dec 202216 Dec 2022

Publication series

Name2022 IEEE International Workshop on Information Forensics and Security, WIFS 2022

Conference

Conference2022 IEEE International Workshop on Information Forensics and Security, WIFS 2022
Country/TerritoryChina
CityShanghai
Period12/12/2216/12/22

Fingerprint

Dive into the research topics of 'Demosaicing to Detect Demosaicing and Image Forgeries'. Together they form a unique fingerprint.

Cite this