Detecting image forgery in single-sensor multispectral images

Abstract

With the advancements in digital technology, multispectral images have found use in fields like forensics, remote sensing due to their ability to perceive things which were otherwise non-existent. They are used to obtain more information about terrains, land cover and in forensics as certain things like blood stains are not visible in visible spectrum. But with newly developed photo-editing softwares, they can be easily manipulated without leaving any visible clue of manipulation, but will destroy the underlying correlation between different bands. Newly developed digital cameras employ a single sensor along with multispectral filter array (MSFA) and then interpolate the data at other locations, hence introducing a correlation between bands. In this paper, we have proposed an algorithm that can identify the lack of correlation at tampered locations in a multispectral image and can thus help in establishing the authenticity of the given multispectral image. We show the efficiency of our approach with respect to the size of tampered regions in images interpolated with one the most common demosaicking algorithm—binary tree-based edge sensing (BTES).