A multi-channel approach for detecting tampering in colour filter images

Abstract

Most devices capable of capturing photographs are provided with a Colour Filter Array (CFA). Images generated by these devices can be analysed to detect the so known CFA artefacts left by demosaicing methods. These artefacts are important in digital image forensics since they prove useful for determining the authenticity of an image. Previous works have focused exclusively on the analysis of the green band of the Bayer filter, neglecting the information in the remaining bands. In this work a process to estimate the pattern left by CFA artefacts regardless of their configuration is proposed, which can be used regardless of the size or colours used by the filter, or even in different colour spaces, obtaining in this way new sources of information for forensic analysis. A review and comparisons against state-of-the-art projects addressing the detection of manipulations using CFA artefacts are performed. The sensitivity of these proposals to post-processing methods such as compression and filtering will also be reviewed. We will also observe experimentally that the proposed technique can detect manipulations even in images affected by high levels of JPEG compression.