CFA demosaicking and its applications in digital image forensics

Today, our understanding of events is highly dependent on visual information, consequently, digital images have become a prominent carrier of information. Dependence of our lifestyles on digital images leads to the critical importance of their security and reliability. In parallel to the technological advancements, availability of software tools capable of easily manipulating digital images in a user-friendly manner is not so rare. With this, the trust we keep on digital images as proof of an event is at risk. This thesis is aimed at contributing to the development of tools for providing authenticity and security to digital images and characterizing imaging devices, particularly digital cameras. The techniques developed in this thesis are based on intrinsic signatures of imaging devices acquired in the image during the image generation process. Most of the cameras available in consumer electronics market, use color filter array (CFA) in conjunction with the sensor array, which introduces some specific statistical signatures in the captured image. When an image generated by such a camera is manipulated, these signatures are destroyed. Hence, the absence of these signatures, indicates that the image is likely to be manipulated. This dissertation presents two passive approaches based on these CFA signatures to detect image forgeries and classify/identify imaging device. The first method is capable of detecting the presence of CFA signatures in a suspect image and is also capable of classifying/identifying imaging device, but is not very robust to lossy compression. The second method presents a solution robust to lossy compression for detecting the presence of CFA de-mosaicking, but is not capable of estimating underlying de-mosaicking scheme. In this way, both the proposed methods complement each other and may be used depending on the application scenario.